Upload folder using huggingface_hub

.eslintignore

@@ -0,0 +1,5 @@
+extensions
+extensions-disabled
+extensions-builtin/sd_forge_controlnet
+repositories
+venv

.eslintrc.js

@@ -0,0 +1,96 @@
+/* global module */
+module.exports = {
+    env: {
+        browser: true,
+        es2021: true,
+    },
+    extends: "eslint:recommended",
+    parserOptions: {
+        ecmaVersion: "latest",
+    },
+    rules: {
+        "arrow-spacing": "error",
+        "block-spacing": "error",
+        "brace-style": "error",
+        "comma-dangle": ["error", "only-multiline"],
+        "comma-spacing": "error",
+        "comma-style": ["error", "last"],
+        "curly": ["error", "multi-line", "consistent"],
+        "eol-last": "error",
+        "func-call-spacing": "error",
+        "function-call-argument-newline": ["error", "consistent"],
+        "function-paren-newline": ["error", "consistent"],
+        "indent": ["error", 4],
+        "key-spacing": "error",
+        "keyword-spacing": "error",
+        "linebreak-style": ["error", "unix"],
+        "no-extra-semi": "error",
+        "no-mixed-spaces-and-tabs": "error",
+        "no-multi-spaces": "error",
+        "no-redeclare": ["error", {builtinGlobals: false}],
+        "no-trailing-spaces": "error",
+        "no-unused-vars": "off",
+        "no-whitespace-before-property": "error",
+        "object-curly-newline": ["error", {consistent: true, multiline: true}],
+        "object-curly-spacing": ["error", "never"],
+        "operator-linebreak": ["error", "after"],
+        "quote-props": ["error", "consistent-as-needed"],
+        "semi": ["error", "always"],
+        "semi-spacing": "error",
+        "semi-style": ["error", "last"],
+        "space-before-blocks": "error",
+        "space-before-function-paren": ["error", "never"],
+        "space-in-parens": ["error", "never"],
+        "space-infix-ops": "error",
+        "space-unary-ops": "error",
+        "switch-colon-spacing": "error",
+        "template-curly-spacing": ["error", "never"],
+        "unicode-bom": "error",
+    },
+    globals: {
+        //script.js
+        gradioApp: "readonly",
+        executeCallbacks: "readonly",
+        onAfterUiUpdate: "readonly",
+        onOptionsChanged: "readonly",
+        onUiLoaded: "readonly",
+        onUiUpdate: "readonly",
+        uiCurrentTab: "writable",
+        uiElementInSight: "readonly",
+        uiElementIsVisible: "readonly",
+        //ui.js
+        opts: "writable",
+        all_gallery_buttons: "readonly",
+        selected_gallery_button: "readonly",
+        selected_gallery_index: "readonly",
+        switch_to_txt2img: "readonly",
+        switch_to_img2img_tab: "readonly",
+        switch_to_img2img: "readonly",
+        switch_to_sketch: "readonly",
+        switch_to_inpaint: "readonly",
+        switch_to_inpaint_sketch: "readonly",
+        switch_to_extras: "readonly",
+        get_tab_index: "readonly",
+        create_submit_args: "readonly",
+        restart_reload: "readonly",
+        updateInput: "readonly",
+        onEdit: "readonly",
+        //extraNetworks.js
+        requestGet: "readonly",
+        popup: "readonly",
+        // from python
+        localization: "readonly",
+        // progrssbar.js
+        randomId: "readonly",
+        requestProgress: "readonly",
+        // imageviewer.js
+        modalPrevImage: "readonly",
+        modalNextImage: "readonly",
+        // localStorage.js
+        localSet: "readonly",
+        localGet: "readonly",
+        localRemove: "readonly",
+        // resizeHandle.js
+        setupResizeHandle: "writable"
+    }
+};

.git-blame-ignore-revs

@@ -0,0 +1,2 @@
+# Apply ESlint
+9c54b78d9dde5601e916f308d9a9d6953ec39430

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -0,0 +1,105 @@
+name: Bug Report
+description: You think something is broken in the UI
+title: "[Bug]: "
+labels: ["bug-report"]
+
+body:
+  - type: markdown
+    attributes:
+      value: |
+        > The title of the bug report should be short and descriptive.
+        > Use relevant keywords for searchability.
+        > Do not leave it blank, but also do not put an entire error log in it.
+  - type: checkboxes
+    attributes:
+      label: Checklist
+      description: |
+        Please perform basic debugging to see if extensions or configuration is the cause of the issue.
+        Basic debug procedure
+        　1. Disable all third-party extensions - check if extension is the cause
+        　2. Update extensions and webui - sometimes things just need to be updated
+        　3. Backup and remove your config.json and ui-config.json - check if the issue is caused by bad configuration
+        　4. Delete venv with third-party extensions disabled - sometimes extensions might cause wrong libraries to be installed
+        　5. Try a fresh installation webui in a different directory - see if a clean installation solves the issue
+        Before making a issue report please, check that the issue hasn't been reported recently.
+      options:
+        - label: The issue exists after disabling all extensions
+        - label: The issue exists on a clean installation of webui
+        - label: The issue is caused by an extension, but I believe it is caused by a bug in the webui
+        - label: The issue exists in the current version of the webui
+        - label: The issue has not been reported before recently
+        - label: The issue has been reported before but has not been fixed yet
+  - type: markdown
+    attributes:
+      value: |
+        > Please fill this form with as much information as possible. Don't forget to "Upload Sysinfo" and "What browsers" and provide screenshots if possible
+  - type: textarea
+    id: what-did
+    attributes:
+      label: What happened?
+      description: Tell us what happened in a very clear and simple way
+      placeholder: |
+        txt2img is not working as intended.
+    validations:
+      required: true
+  - type: textarea
+    id: steps
+    attributes:
+      label: Steps to reproduce the problem
+      description: Please provide us with precise step by step instructions on how to reproduce the bug
+      placeholder: |
+        1. Go to ...
+        2. Press ...
+        3. ...
+    validations:
+      required: true
+  - type: textarea
+    id: what-should
+    attributes:
+      label: What should have happened?
+      description: Tell us what you think the normal behavior should be
+      placeholder: |
+        WebUI should ...
+    validations:
+      required: true
+  - type: dropdown
+    id: browsers
+    attributes:
+      label: What browsers do you use to access the UI ?
+      multiple: true
+      options:
+        - Mozilla Firefox
+        - Google Chrome
+        - Brave
+        - Apple Safari
+        - Microsoft Edge
+        - Android
+        - iOS
+        - Other
+  - type: textarea
+    id: sysinfo
+    attributes:
+      label: Sysinfo
+      description: System info file, generated by WebUI. You can generate it in settings, on the Sysinfo page. Drag the file into the field to upload it. If you submit your report without including the sysinfo file, the report will be closed. If needed, review the report to make sure it includes no personal information you don't want to share. If you can't start WebUI, you can use --dump-sysinfo commandline argument to generate the file.
+      placeholder: |
+        1. Go to WebUI Settings -> Sysinfo -> Download system info.
+            If WebUI fails to launch, use --dump-sysinfo commandline argument to generate the file
+        2. Upload the Sysinfo as a attached file, Do NOT paste it in as plain text.
+    validations:
+      required: true
+  - type: textarea
+    id: logs
+    attributes:
+      label: Console logs
+      description: Please provide **full** cmd/terminal logs from the moment you started UI to the end of it, after the bug occured. If it's very long, provide a link to pastebin or similar service.
+      render: Shell
+    validations:
+      required: true
+  - type: textarea
+    id: misc
+    attributes:
+      label: Additional information
+      description: | 
+        Please provide us with any relevant additional info or context.
+        Examples:
+        　I have updated my GPU driver recently.

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,5 @@
+blank_issues_enabled: false
+contact_links:
+  - name: WebUI Community Support
+    url: https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions
+    about: Please ask and answer questions here.

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -0,0 +1,40 @@
+name: Feature request
+description: Suggest an idea for this project
+title: "[Feature Request]: "
+labels: ["enhancement"]
+
+body:
+  - type: checkboxes
+    attributes:
+      label: Is there an existing issue for this?
+      description: Please search to see if an issue already exists for the feature you want, and that it's not implemented in a recent build/commit.
+      options:
+        - label: I have searched the existing issues and checked the recent builds/commits
+          required: true
+  - type: markdown
+    attributes:
+      value: |
+        *Please fill this form with as much information as possible, provide screenshots and/or illustrations of the feature if possible*
+  - type: textarea
+    id: feature
+    attributes:
+      label: What would your feature do ?
+      description: Tell us about your feature in a very clear and simple way, and what problem it would solve
+    validations:
+      required: true
+  - type: textarea
+    id: workflow
+    attributes:
+      label: Proposed workflow
+      description: Please provide us with step by step information on how you'd like the feature to be accessed and used
+      value: |
+        1. Go to .... 
+        2. Press ....
+        3. ...
+    validations:
+      required: true
+  - type: textarea
+    id: misc
+    attributes:
+      label: Additional information
+      description: Add any other context or screenshots about the feature request here.

.github/pull_request_template.md

@@ -0,0 +1,15 @@
+## Description
+
+* a simple description of what you're trying to accomplish
+* a summary of changes in code
+* which issues it fixes, if any
+
+## Screenshots/videos:
+
+
+## Checklist:
+
+- [ ] I have read [contributing wiki page](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Contributing)
+- [ ] I have performed a self-review of my own code
+- [ ] My code follows the [style guidelines](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Contributing#code-style)
+- [ ] My code passes [tests](https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Tests)

.github/workflows/on_pull_request.yaml

@@ -0,0 +1,38 @@
+name: Linter
+
+on:
+  - push
+  - pull_request
+
+jobs:
+  lint-python:
+    name: ruff
+    runs-on: ubuntu-latest
+    if: github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name != github.event.pull_request.base.repo.full_name
+    steps:
+      - name: Checkout Code
+        uses: actions/checkout@v3
+      - uses: actions/setup-python@v4
+        with:
+          python-version: 3.11
+          # NB: there's no cache: pip here since we're not installing anything
+          #     from the requirements.txt file(s) in the repository; it's faster
+          #     not to have GHA download an (at the time of writing) 4 GB cache
+          #     of PyTorch and other dependencies.
+      - name: Install Ruff
+        run: pip install ruff==0.1.6
+      - name: Run Ruff
+        run: ruff .
+  lint-js:
+    name: eslint
+    runs-on: ubuntu-latest
+    if: github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name != github.event.pull_request.base.repo.full_name
+    steps:
+      - name: Checkout Code
+        uses: actions/checkout@v3
+      - name: Install Node.js
+        uses: actions/setup-node@v3
+        with:
+          node-version: 18
+      - run: npm i --ci
+      - run: npm run lint

.github/workflows/run_tests.yaml

@@ -0,0 +1,107 @@
+name: Tests
+
+on:
+  - push
+  - pull_request
+
+env:
+  FORGE_CQ_TEST: "True"
+
+jobs:
+  test:
+    name: tests on CPU
+    runs-on: ubuntu-latest
+    if: github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name != github.event.pull_request.base.repo.full_name
+    steps:
+      - name: Checkout Code
+        uses: actions/checkout@v3
+      - name: Set up Python 3.10
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.10.6
+          cache: pip
+          cache-dependency-path: |
+            **/requirements*txt
+            launch.py
+      - name: Cache models
+        id: cache-models
+        uses: actions/cache@v3
+        with:
+          path: models
+          key: "2023-12-30"
+      - name: Install test dependencies
+        run: pip install wait-for-it -r requirements-test.txt
+        env:
+          PIP_DISABLE_PIP_VERSION_CHECK: "1"
+          PIP_PROGRESS_BAR: "off"
+      - name: Setup environment
+        run: python launch.py --skip-torch-cuda-test --exit
+        env:
+          PIP_DISABLE_PIP_VERSION_CHECK: "1"
+          PIP_PROGRESS_BAR: "off"
+          TORCH_INDEX_URL: https://download.pytorch.org/whl/cpu
+          WEBUI_LAUNCH_LIVE_OUTPUT: "1"
+          PYTHONUNBUFFERED: "1"
+      - name: Print installed packages
+        run: pip freeze
+      - name: Download models
+        run: |
+          declare -a urls=(
+            "https://huggingface.co/lllyasviel/fav_models/resolve/main/fav/realisticVisionV51_v51VAE.safetensors"
+          )
+          for url in "${urls[@]}"; do
+            filename="models/Stable-diffusion/${url##*/}"  # Extracts the last part of the URL
+            if [ ! -f "$filename" ]; then
+              curl -Lo "$filename" "$url"
+            fi
+          done
+      # - name: Download ControlNet models
+      #   run: |
+      #     declare -a urls=(
+      #       "https://huggingface.co/lllyasviel/ControlNet-v1-1/resolve/main/control_v11p_sd15_canny.pth"
+      #     )
+
+      #     for url in "${urls[@]}"; do
+      #       filename="models/ControlNet/${url##*/}"  # Extracts the last part of the URL
+      #       if [ ! -f "$filename" ]; then
+      #         curl -Lo "$filename" "$url"
+      #       fi
+      #     done
+      - name: Start test server
+        run: >
+          python -m coverage run
+          --data-file=.coverage.server
+          launch.py
+          --skip-prepare-environment
+          --skip-torch-cuda-test
+          --test-server
+          --do-not-download-clip
+          --no-half
+          --disable-opt-split-attention
+          --always-cpu
+          --api-server-stop
+          --ckpt models/Stable-diffusion/realisticVisionV51_v51VAE.safetensors
+          2>&1 | tee output.txt &
+      - name: Run tests
+        run: |
+          wait-for-it --service 127.0.0.1:7860 -t 20
+          python -m pytest -vv --junitxml=test/results.xml --cov . --cov-report=xml --verify-base-url test
+      # TODO(huchenlei): Enable ControlNet tests. Currently it is too slow to run these tests on CPU with
+      # real SD model. We need to find a way to load empty SD model.
+      # - name: Run ControlNet tests
+      #   run: >
+      #     python -m pytest
+      #     --junitxml=test/results.xml
+      #     --cov ./extensions-builtin/sd_forge_controlnet
+      #     --cov-report=xml
+      #     --verify-base-url
+      #     ./extensions-builtin/sd_forge_controlnet/tests
+      - name: Kill test server
+        if: always()
+        run: curl -vv -XPOST http://127.0.0.1:7860/sdapi/v1/server-stop && sleep 10
+      - name: Upload main app output
+        uses: actions/upload-artifact@v3
+        if: always()
+        with:
+          name: output
+          path: output.txt

.github/workflows/warns_merge_master.yml

@@ -0,0 +1,19 @@
+name: Pull requests can't target master branch
+
+"on":
+  pull_request:
+    types:
+      - opened
+      - synchronize
+      - reopened
+    branches:
+      - master
+
+jobs:
+  check:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Warning marge into master
+        run: |
+          echo -e "::warning::This pull request directly merge into \"master\" branch, normally development happens on \"dev\" branch."
+          exit 1

.gitignore

@@ -0,0 +1,44 @@
+__pycache__
+*.ckpt
+*.safetensors
+*.pth
+/ESRGAN/*
+/SwinIR/*
+/repositories
+/venv
+/tmp
+/output
+/model.ckpt
+/models/**/*
+/GFPGANv1.3.pth
+/gfpgan/weights/*.pth
+/ui-config.json
+/outputs
+/config.json
+/log
+/webui.settings.bat
+/embeddings
+/styles.csv
+/params.txt
+/styles.csv.bak
+/webui-user.bat
+/webui-user.sh
+/interrogate
+/user.css
+/.idea
+notification.mp3
+/SwinIR
+/textual_inversion
+.vscode
+/extensions
+/test/stdout.txt
+/test/stderr.txt
+/cache.json*
+/config_states/
+/node_modules
+/package-lock.json
+/.coverage*
+/test/test_outputs
+/test/results.xml
+coverage.xml
+**/tests/**/expectations

.pylintrc

@@ -0,0 +1,3 @@
+# See https://pylint.pycqa.org/en/latest/user_guide/messages/message_control.html
+[MESSAGES CONTROL]
+disable=C,R,W,E,I

CHANGELOG.md

@@ -0,0 +1,674 @@
+## 1.7.0
+
+### Features:
+* settings tab rework: add search field, add categories, split UI settings page into many
+* add altdiffusion-m18 support ([#13364](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13364))
+* support inference with LyCORIS GLora networks ([#13610](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13610))
+* add lora-embedding bundle system ([#13568](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13568))
+* option to move prompt from top row into generation parameters
+* add support for SSD-1B ([#13865](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13865))
+* support inference with OFT networks ([#13692](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13692))
+* script metadata and DAG sorting mechanism ([#13944](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13944))
+* support HyperTile optimization ([#13948](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13948))
+* add support for SD 2.1 Turbo ([#14170](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14170))
+* remove Train->Preprocessing tab and put all its functionality into Extras tab
+* initial IPEX support for Intel Arc GPU ([#14171](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14171))
+
+### Minor:
+* allow reading model hash from images in img2img batch mode ([#12767](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12767))
+* add option to align with sgm repo's sampling implementation ([#12818](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12818))
+* extra field for lora metadata viewer: `ss_output_name` ([#12838](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12838))
+* add action in settings page to calculate all SD checkpoint hashes ([#12909](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12909))
+* add button to copy prompt to style editor ([#12975](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12975))
+* add --skip-load-model-at-start option ([#13253](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13253))
+* write infotext to gif images
+* read infotext from gif images ([#13068](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13068))
+* allow configuring the initial state of InputAccordion in ui-config.json ([#13189](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13189))
+* allow editing whitespace delimiters for ctrl+up/ctrl+down prompt editing ([#13444](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13444))
+* prevent accidentally closing popup dialogs ([#13480](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13480))
+* added option to play notification sound or not ([#13631](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13631))
+* show the preview image in the full screen image viewer if available ([#13459](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13459))
+* support for webui.settings.bat ([#13638](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13638))
+* add an option to not print stack traces on ctrl+c
+* start/restart generation by Ctrl (Alt) + Enter ([#13644](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13644))
+* update prompts_from_file script to allow concatenating entries with the general prompt ([#13733](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13733))
+* added a visible checkbox to input accordion
+* added an option to hide all txt2img/img2img parameters in an accordion ([#13826](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13826))
+* added 'Path' sorting option for Extra network cards ([#13968](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13968))
+* enable prompt hotkeys in style editor ([#13931](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13931))
+* option to show batch img2img results in UI ([#14009](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14009))
+* infotext updates: add option to disregard certain infotext fields, add option to not include VAE in infotext, add explanation to infotext settings page, move some options to infotext settings page
+* add FP32 fallback support on sd_vae_approx ([#14046](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14046))
+* support XYZ scripts / split hires path from unet ([#14126](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14126))
+* allow use of mutiple styles csv files ([#14125](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14125))
+
+### Extensions and API:
+* update gradio to 3.41.2
+* support installed extensions list api ([#12774](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12774))
+* update pnginfo API to return dict with parsed values
+* add noisy latent to `ExtraNoiseParams` for callback ([#12856](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12856))
+* show extension datetime in UTC ([#12864](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12864), [#12865](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12865), [#13281](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13281))
+* add an option to choose how to combine hires fix and refiner
+* include program version in info response. ([#13135](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13135))
+* sd_unet support for SDXL
+* patch DDPM.register_betas so that users can put given_betas in model yaml ([#13276](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13276))
+* xyz_grid: add prepare ([#13266](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13266))
+* allow multiple localization files with same language in extensions ([#13077](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13077))
+* add onEdit function for js and rework token-counter.js to use it
+* fix the key error exception when processing override_settings keys ([#13567](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13567))
+* ability for extensions to return custom data via api in response.images ([#13463](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13463))
+* call state.jobnext() before postproces*() ([#13762](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13762))
+* add option to set notification sound volume ([#13884](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13884))
+* update Ruff to 0.1.6 ([#14059](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14059))
+* add Block component creation callback ([#14119](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14119))
+* catch uncaught exception with ui creation scripts ([#14120](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14120))
+* use extension name for determining an extension is installed in the index ([#14063](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14063))
+* update is_installed() from launch_utils.py to fix reinstalling already installed packages ([#14192](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14192))
+
+### Bug Fixes:
+* fix pix2pix producing bad results
+* fix defaults settings page breaking when any of main UI tabs are hidden
+* fix error that causes some extra networks to be disabled if both <lora:> and <lyco:> are present in the prompt
+* fix for Reload UI function: if you reload UI on one tab, other opened tabs will no longer stop working
+* prevent duplicate resize handler ([#12795](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12795))
+* small typo: vae resolve bug ([#12797](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12797))
+* hide broken image crop tool ([#12792](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12792))
+* don't show hidden samplers in dropdown for XYZ script ([#12780](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12780))
+* fix style editing dialog breaking if it's opened in both img2img and txt2img tabs
+* hide --gradio-auth and --api-auth values from /internal/sysinfo report
+* add missing infotext for RNG in options ([#12819](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12819))
+* fix notification not playing when built-in webui tab is inactive ([#12834](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12834))
+* honor `--skip-install` for extension installers ([#12832](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12832))
+* don't print blank stdout in extension installers ([#12833](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12833), [#12855](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12855))
+* get progressbar to display correctly in extensions tab
+* keep order in list of checkpoints when loading model that doesn't have a checksum
+* fix inpainting models in txt2img creating black pictures
+* fix generation params regex ([#12876](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12876))
+* fix batch img2img output dir with script ([#12926](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12926))
+* fix #13080 - Hypernetwork/TI preview generation ([#13084](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13084))
+* fix bug with sigma min/max overrides. ([#12995](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12995))
+* more accurate check for enabling cuDNN benchmark on 16XX cards ([#12924](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12924))
+* don't use multicond parser for negative prompt counter ([#13118](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13118))
+* fix data-sort-name containing spaces ([#13412](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13412))
+* update card on correct tab when editing metadata ([#13411](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13411))
+* fix viewing/editing metadata when filename contains an apostrophe ([#13395](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13395))
+* fix: --sd_model in "Prompts from file or textbox" script is not working ([#13302](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13302))
+* better Support for Portable Git ([#13231](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13231))
+* fix issues when webui_dir is not work_dir ([#13210](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13210))
+* fix: lora-bias-backup don't reset cache ([#13178](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13178))
+* account for customizable extra network separators whyen removing extra network text from the prompt ([#12877](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12877))
+* re fix batch img2img output dir with script ([#13170](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13170))
+* fix `--ckpt-dir` path separator and option use `short name` for checkpoint dropdown ([#13139](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13139))
+* consolidated allowed preview formats, Fix extra network `.gif` not woking as preview ([#13121](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13121))
+* fix venv_dir=- environment variable not working as expected on linux ([#13469](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13469))
+* repair unload sd checkpoint button
+* edit-attention fixes ([#13533](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13533))
+* fix bug when using --gfpgan-models-path ([#13718](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13718))
+* properly apply sort order for extra network cards when selected from dropdown
+* fixes generation restart not working for some users when 'Ctrl+Enter' is pressed ([#13962](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13962))
+* thread safe extra network list_items ([#13014](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13014))
+* fix not able to exit metadata popup when pop up is too big ([#14156](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14156))
+* fix auto focal point crop for opencv >= 4.8 ([#14121](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14121))
+* make 'use-cpu all' actually apply to 'all' ([#14131](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14131))
+* extras tab batch: actually use original filename
+* make webui not crash when running with --disable-all-extensions option
+
+### Other:
+* non-local condition ([#12814](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12814))
+* fix minor typos ([#12827](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12827))
+* remove xformers Python version check ([#12842](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12842))
+* style: file-metadata word-break ([#12837](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12837))
+* revert SGM noise multiplier change for img2img because it breaks hires fix
+* do not change quicksettings dropdown option when value returned is `None` ([#12854](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12854))
+* [RC 1.6.0 - zoom is partly hidden] Update style.css ([#12839](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12839))
+* chore: change extension time format ([#12851](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12851))
+* WEBUI.SH - Use torch 2.1.0 release candidate for Navi 3 ([#12929](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12929))
+* add Fallback at images.read_info_from_image if exif data was invalid ([#13028](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13028))
+* update cmd arg description ([#12986](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12986))
+* fix: update shared.opts.data when add_option ([#12957](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12957), [#13213](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13213))
+* restore missing tooltips ([#12976](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12976))
+* use default dropdown padding on mobile ([#12880](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12880))
+* put enable console prompts option into settings from commandline args ([#13119](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13119))
+* fix some deprecated types ([#12846](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12846))
+* bump to torchsde==0.2.6 ([#13418](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13418))
+* update dragdrop.js ([#13372](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13372))
+* use orderdict as lru cache:opt/bug ([#13313](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13313))
+* XYZ if not include sub grids do not save sub grid ([#13282](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13282))
+* initialize state.time_start befroe state.job_count ([#13229](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13229))
+* fix fieldname regex ([#13458](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13458))
+* change denoising_strength default to None. ([#13466](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13466))
+* fix regression ([#13475](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13475))
+* fix IndexError ([#13630](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13630))
+* fix: checkpoints_loaded:{checkpoint:state_dict}, model.load_state_dict issue in dict value empty ([#13535](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13535))
+* update bug_report.yml ([#12991](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12991))
+* requirements_versions httpx==0.24.1 ([#13839](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13839))
+* fix parenthesis auto selection ([#13829](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13829))
+* fix #13796 ([#13797](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13797))
+* corrected a typo in `modules/cmd_args.py` ([#13855](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13855))
+* feat: fix randn found element of type float at pos 2 ([#14004](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14004))
+* adds tqdm handler to logging_config.py for progress bar integration ([#13996](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13996))
+* hotfix: call shared.state.end() after postprocessing done ([#13977](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13977))
+* fix dependency address patch 1 ([#13929](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13929))
+* save sysinfo as .json ([#14035](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14035))
+* move exception_records related methods to errors.py ([#14084](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14084))
+* compatibility ([#13936](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13936))
+* json.dump(ensure_ascii=False) ([#14108](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14108))
+* dir buttons start with / so only the correct dir will be shown and no… ([#13957](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13957))
+* alternate implementation for unet forward replacement that does not depend on hijack being applied
+* re-add `keyedit_delimiters_whitespace` setting lost as part of commit e294e46 ([#14178](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14178))
+* fix `save_samples` being checked early when saving masked composite ([#14177](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14177))
+* slight optimization for mask and mask_composite ([#14181](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14181))
+* add import_hook hack to work around basicsr/torchvision incompatibility ([#14186](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14186))
+
+## 1.6.1
+
+### Bug Fixes:
+ * fix an error causing the webui to fail to start ([#13839](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13839))
+
+## 1.6.0
+
+### Features:
+ * refiner support [#12371](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12371)
+ * add NV option for Random number generator source setting, which allows to generate same pictures on CPU/AMD/Mac as on NVidia videocards
+ * add style editor dialog
+ * hires fix: add an option to use a different checkpoint for second pass ([#12181](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12181))
+ * option to keep multiple loaded models in memory ([#12227](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12227))
+ * new samplers: Restart, DPM++ 2M SDE Exponential, DPM++ 2M SDE Heun, DPM++ 2M SDE Heun Karras, DPM++ 2M SDE Heun Exponential, DPM++ 3M SDE, DPM++ 3M SDE Karras, DPM++ 3M SDE Exponential ([#12300](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12300), [#12519](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12519), [#12542](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12542))
+ * rework DDIM, PLMS, UniPC to use CFG denoiser same as in k-diffusion samplers:
+   * makes all of them work with img2img
+   * makes prompt composition posssible (AND)
+   * makes them available for SDXL
+ * always show extra networks tabs in the UI ([#11808](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/11808))
+ * use less RAM when creating models ([#11958](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/11958), [#12599](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12599))
+ * textual inversion inference support for SDXL
+ * extra networks UI: show metadata for SD checkpoints
+ * checkpoint merger: add metadata support 
+ * prompt editing and attention: add support for whitespace after the number ([ red : green : 0.5 ]) (seed breaking change) ([#12177](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12177))
+ * VAE: allow selecting own VAE for each checkpoint (in user metadata editor)
+ * VAE: add selected VAE to infotext
+ * options in main UI: add own separate setting for txt2img and img2img, correctly read values from pasted infotext, add setting for column count ([#12551](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12551))
+ * add resize handle to txt2img and img2img tabs, allowing to change the amount of horizontable space given to generation parameters and resulting image gallery ([#12687](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12687), [#12723](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12723))
+ * change default behavior for batching cond/uncond -- now it's on by default, and is disabled by an UI setting (Optimizatios -> Batch cond/uncond) - if you are on lowvram/medvram and are getting OOM exceptions, you will need to enable it
+ * show current position in queue and make it so that requests are processed in the order of arrival ([#12707](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12707))
+ * add `--medvram-sdxl` flag that only enables `--medvram` for SDXL models
+ * prompt editing timeline has separate range for first pass and hires-fix pass (seed breaking change) ([#12457](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12457))
+
+### Minor:
+ * img2img batch: RAM savings, VRAM savings, .tif, .tiff in img2img batch ([#12120](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12120), [#12514](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12514), [#12515](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12515))
+ * postprocessing/extras: RAM savings ([#12479](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12479))
+ * XYZ: in the axis labels, remove pathnames from model filenames
+ * XYZ: support hires sampler ([#12298](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12298))
+ * XYZ: new option: use text inputs instead of dropdowns ([#12491](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12491))
+ * add gradio version warning
+ * sort list of VAE checkpoints ([#12297](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12297))
+ * use transparent white for mask in inpainting, along with an option to select the color ([#12326](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12326))
+ * move some settings to their own section: img2img, VAE
+ * add checkbox to show/hide dirs for extra networks
+ * Add TAESD(or more) options for all the VAE encode/decode operation ([#12311](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12311))
+ * gradio theme cache, new gradio themes, along with explanation that the user can input his own values ([#12346](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12346), [#12355](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12355))
+ * sampler fixes/tweaks: s_tmax, s_churn, s_noise, s_tmax ([#12354](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12354), [#12356](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12356), [#12357](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12357), [#12358](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12358), [#12375](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12375), [#12521](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12521))
+ * update README.md with correct instructions for Linux installation ([#12352](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12352))
+ * option to not save incomplete images, on by default ([#12338](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12338))
+ * enable cond cache by default
+ * git autofix for repos that are corrupted ([#12230](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12230))
+ * allow to open images in new browser tab by middle mouse button ([#12379](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12379))
+ * automatically open webui in browser when running "locally" ([#12254](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12254))
+ * put commonly used samplers on top, make DPM++ 2M Karras the default choice
+ * zoom and pan: option to auto-expand a wide image, improved integration ([#12413](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12413), [#12727](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12727))
+ * option to cache Lora networks in memory
+ * rework hires fix UI to use accordion
+ * face restoration and tiling moved to settings - use "Options in main UI" setting if you want them back
+ * change quicksettings items to have variable width
+ * Lora: add Norm module, add support for bias ([#12503](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12503))
+ * Lora: output warnings in UI rather than fail for unfitting loras; switch to logging for error output in console
+ * support search and display of hashes for all extra network items ([#12510](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12510))
+ * add extra noise param for img2img operations ([#12564](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12564))
+ * support for Lora with bias ([#12584](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12584))
+ * make interrupt quicker ([#12634](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12634))
+ * configurable gallery height ([#12648](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12648))
+ * make results column sticky ([#12645](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12645))
+ * more hash filename patterns ([#12639](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12639))
+ * make image viewer actually fit the whole page ([#12635](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12635))
+ * make progress bar work independently from live preview display which results in it being updated a lot more often
+ * forbid Full live preview method for medvram and add a setting to undo the forbidding
+ * make it possible to localize tooltips and placeholders
+ * add option to align with sgm repo's sampling implementation ([#12818](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12818))
+ * Restore faces and Tiling generation parameters have been moved to settings out of main UI
+   * if you want to put them back into main UI, use `Options in main UI` setting on the UI page.
+
+### Extensions and API:
+ * gradio 3.41.2
+ * also bump versions for packages: transformers, GitPython, accelerate, scikit-image, timm, tomesd
+ * support tooltip kwarg for gradio elements: gr.Textbox(label='hello', tooltip='world')
+ * properly clear the total console progressbar when using txt2img and img2img from API
+ * add cmd_arg --disable-extra-extensions and --disable-all-extensions ([#12294](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12294))
+ * shared.py and webui.py split into many files
+ * add --loglevel commandline argument for logging
+ * add a custom UI element that combines accordion and checkbox
+ * avoid importing gradio in tests because it spams warnings
+ * put infotext label for setting into OptionInfo definition rather than in a separate list
+ * make `StableDiffusionProcessingImg2Img.mask_blur` a property, make more inline with PIL `GaussianBlur` ([#12470](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12470))
+ * option to make scripts UI without gr.Group
+ * add a way for scripts to register a callback for before/after just a single component's creation
+ * use dataclass for StableDiffusionProcessing
+ * store patches for Lora in a specialized module instead of inside torch
+ * support http/https URLs in API ([#12663](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12663), [#12698](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12698))
+ * add extra noise callback ([#12616](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12616))
+ * dump current stack traces when exiting with SIGINT
+ * add type annotations for extra fields of shared.sd_model
+
+### Bug Fixes:
+ * Don't crash if out of local storage quota for javascriot localStorage
+ * XYZ plot do not fail if an exception occurs
+ * fix missing TI hash in infotext if generation uses both negative and positive TI ([#12269](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12269))
+ * localization fixes ([#12307](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12307))
+ * fix sdxl model invalid configuration after the hijack
+ * correctly toggle extras checkbox for infotext paste ([#12304](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12304))
+ * open raw sysinfo link in new page ([#12318](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12318))
+ * prompt parser: Account for empty field in alternating words syntax ([#12319](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12319))
+ * add tab and carriage return to invalid filename chars ([#12327](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12327))
+ * fix api only Lora not working ([#12387](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12387))
+ * fix options in main UI misbehaving when there's just one element
+ * make it possible to use a sampler from infotext even if it's hidden in the dropdown
+ * fix styles missing from the prompt in infotext when making a grid of batch of multiplie images
+ * prevent bogus progress output in console when calculating hires fix dimensions
+ * fix --use-textbox-seed
+ * fix broken `Lora/Networks: use old method` option ([#12466](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12466))
+ * properly return `None` for VAE hash when using `--no-hashing` ([#12463](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12463))
+ * MPS/macOS fixes and optimizations ([#12526](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12526))
+ * add second_order to samplers that mistakenly didn't have it
+ * when refreshing cards in extra networks UI, do not discard user's custom resolution
+ * fix processing error that happens if batch_size is not a multiple of how many prompts/negative prompts there are ([#12509](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12509))
+ * fix inpaint upload for alpha masks ([#12588](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12588))
+ * fix exception when image sizes are not integers ([#12586](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12586))
+ * fix incorrect TAESD Latent scale ([#12596](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12596))
+ * auto add data-dir to gradio-allowed-path ([#12603](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12603))
+ * fix exception if extensuions dir is missing ([#12607](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12607))
+ * fix issues with api model-refresh and vae-refresh ([#12638](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12638))
+ * fix img2img background color for transparent images option not being used ([#12633](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12633))
+ * attempt to resolve NaN issue with unstable VAEs in fp32 mk2 ([#12630](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12630))
+ * implement missing undo hijack for SDXL
+ * fix xyz swap axes ([#12684](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12684))
+ * fix errors in backup/restore tab if any of config files are broken ([#12689](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12689))
+ * fix SD VAE switch error after model reuse ([#12685](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12685))
+ * fix trying to create images too large for the chosen format ([#12667](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12667))
+ * create Gradio temp directory if necessary ([#12717](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12717))
+ * prevent possible cache loss if exiting as it's being written by using an atomic operation to replace the cache with the new version
+ * set devices.dtype_unet correctly
+ * run RealESRGAN on GPU for non-CUDA devices ([#12737](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12737))
+ * prevent extra network buttons being obscured by description for very small card sizes ([#12745](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12745))
+ * fix error that causes some extra networks to be disabled if both <lora:> and <lyco:> are present in the prompt
+ * fix defaults settings page breaking when any of main UI tabs are hidden
+ * fix incorrect save/display of new values in Defaults page in settings
+ * fix for Reload UI function: if you reload UI on one tab, other opened tabs will no longer stop working
+ * fix an error that prevents VAE being reloaded after an option change if a VAE near the checkpoint exists ([#12797](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12737))
+ * hide broken image crop tool ([#12792](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12737))
+ * don't show hidden samplers in dropdown for XYZ script ([#12780](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12737))
+ * fix style editing dialog breaking if it's opened in both img2img and txt2img tabs
+ * fix a bug allowing users to bypass gradio and API authentication (reported by vysecurity) 
+ * fix notification not playing when built-in webui tab is inactive ([#12834](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12834))
+ * honor `--skip-install` for extension installers ([#12832](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12832))
+ * don't print blank stdout in extension installers ([#12833](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12832), [#12855](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12855))
+ * do not change quicksettings dropdown option when value returned is `None` ([#12854](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12854))
+ * get progressbar to display correctly in extensions tab
+
+
+## 1.5.2
+
+### Bug Fixes:
+ * fix memory leak when generation fails
+ * update doggettx cross attention optimization to not use an unreasonable amount of memory in some edge cases -- suggestion by MorkTheOrk
+
+
+## 1.5.1
+
+### Minor:
+ * support parsing text encoder blocks in some new LoRAs
+ * delete scale checker script due to user demand
+
+### Extensions and API:
+ * add postprocess_batch_list script callback
+
+### Bug Fixes:
+ * fix TI training for SD1
+ * fix reload altclip model error
+ * prepend the pythonpath instead of overriding it
+ * fix typo in SD_WEBUI_RESTARTING
+ * if txt2img/img2img raises an exception, finally call state.end()
+ * fix composable diffusion weight parsing
+ * restyle Startup profile for black users
+ * fix webui not launching with --nowebui
+ * catch exception for non git extensions
+ * fix some options missing from /sdapi/v1/options
+ * fix for extension update status always saying "unknown"
+ * fix display of extra network cards that have `<>` in the name
+ * update lora extension to work with python 3.8
+
+
+## 1.5.0
+
+### Features:
+ * SD XL support
+ * user metadata system for custom networks
+ * extended Lora metadata editor: set activation text, default weight, view tags, training info
+ * Lora extension rework to include other types of networks (all that were previously handled by LyCORIS extension)
+ * show github stars for extenstions
+ * img2img batch mode can read extra stuff from png info
+ * img2img batch works with subdirectories
+ * hotkeys to move prompt elements: alt+left/right
+ * restyle time taken/VRAM display
+ * add textual inversion hashes to infotext
+ * optimization: cache git extension repo information
+ * move generate button next to the generated picture for mobile clients
+ * hide cards for networks of incompatible Stable Diffusion version in Lora extra networks interface
+ * skip installing packages with pip if they all are already installed - startup speedup of about 2 seconds
+
+### Minor:
+ * checkbox to check/uncheck all extensions in the Installed tab
+ * add gradio user to infotext and to filename patterns
+ * allow gif for extra network previews
+ * add options to change colors in grid
+ * use natural sort for items in extra networks
+ * Mac: use empty_cache() from torch 2 to clear VRAM
+ * added automatic support for installing the right libraries for Navi3 (AMD)
+ * add option SWIN_torch_compile to accelerate SwinIR upscale
+ * suppress printing TI embedding info at start to console by default
+ * speedup extra networks listing
+ * added `[none]` filename token.
+ * removed thumbs extra networks view mode (use settings tab to change width/height/scale to get thumbs)
+ * add always_discard_next_to_last_sigma option to XYZ plot
+ * automatically switch to 32-bit float VAE if the generated picture has NaNs without the need for `--no-half-vae` commandline flag.
+ 
+### Extensions and API:
+ * api endpoints: /sdapi/v1/server-kill, /sdapi/v1/server-restart, /sdapi/v1/server-stop
+ * allow Script to have custom metaclass
+ * add model exists status check /sdapi/v1/options
+ * rename --add-stop-route to --api-server-stop
+ * add `before_hr` script callback
+ * add callback `after_extra_networks_activate`
+ * disable rich exception output in console for API by default, use WEBUI_RICH_EXCEPTIONS env var to enable
+ * return http 404 when thumb file not found
+ * allow replacing extensions index with environment variable
+ 
+### Bug Fixes:
+ * fix for catch errors when retrieving extension index #11290
+ * fix very slow loading speed of .safetensors files when reading from network drives
+ * API cache cleanup
+ * fix UnicodeEncodeError when writing to file CLIP Interrogator batch mode
+ * fix warning of 'has_mps' deprecated from PyTorch
+ * fix problem with extra network saving images as previews losing generation info
+ * fix throwing exception when trying to resize image with I;16 mode
+ * fix for #11534: canvas zoom and pan extension hijacking shortcut keys
+ * fixed launch script to be runnable from any directory
+ * don't add "Seed Resize: -1x-1" to API image metadata
+ * correctly remove end parenthesis with ctrl+up/down
+ * fixing --subpath on newer gradio version
+ * fix: check fill size none zero when resize  (fixes #11425)
+ * use submit and blur for quick settings textbox
+ * save img2img batch with images.save_image()
+ * prevent running preload.py for disabled extensions
+ * fix: previously, model name was added together with directory name to infotext and to [model_name] filename pattern; directory name is now not included
+
+
+## 1.4.1
+
+### Bug Fixes:
+ * add queue lock for refresh-checkpoints
+
+## 1.4.0
+
+### Features:
+ * zoom controls for inpainting
+ * run basic torch calculation at startup in parallel to reduce the performance impact of first generation
+ * option to pad prompt/neg prompt to be same length
+ * remove taming_transformers dependency
+ * custom k-diffusion scheduler settings
+ * add an option to show selected settings in main txt2img/img2img UI
+ * sysinfo tab in settings
+ * infer styles from prompts when pasting params into the UI
+ * an option to control the behavior of the above
+
+### Minor:
+ * bump Gradio to 3.32.0
+ * bump xformers to 0.0.20
+ * Add option to disable token counters
+ * tooltip fixes & optimizations
+ * make it possible to configure filename for the zip download
+ * `[vae_filename]` pattern for filenames
+ * Revert discarding penultimate sigma for DPM-Solver++(2M) SDE
+ * change UI reorder setting to multiselect
+ * read version info form CHANGELOG.md if git version info is not available
+ * link footer API to Wiki when API is not active
+ * persistent conds cache (opt-in optimization)
+ 
+### Extensions:
+ * After installing extensions, webui properly restarts the process rather than reloads the UI 
+ * Added VAE listing to web API. Via: /sdapi/v1/sd-vae
+ * custom unet support
+ * Add onAfterUiUpdate callback
+ * refactor EmbeddingDatabase.register_embedding() to allow unregistering
+ * add before_process callback for scripts
+ * add ability for alwayson scripts to specify section and let user reorder those sections
+ 
+### Bug Fixes:
+ * Fix dragging text to prompt
+ * fix incorrect quoting for infotext values with colon in them
+ * fix "hires. fix" prompt sharing same labels with txt2img_prompt
+ * Fix s_min_uncond default type int
+ * Fix for #10643 (Inpainting mask sometimes not working)
+ * fix bad styling for thumbs view in extra networks #10639
+ * fix for empty list of optimizations #10605
+ * small fixes to prepare_tcmalloc for Debian/Ubuntu compatibility
+ * fix --ui-debug-mode exit
+ * patch GitPython to not use leaky persistent processes
+ * fix duplicate Cross attention optimization after UI reload
+ * torch.cuda.is_available() check for SdOptimizationXformers
+ * fix hires fix using wrong conds in second pass if using Loras.
+ * handle exception when parsing generation parameters from png info
+ * fix upcast attention dtype error
+ * forcing Torch Version to 1.13.1 for RX 5000 series GPUs
+ * split mask blur into X and Y components, patch Outpainting MK2 accordingly
+ * don't die when a LoRA is a broken symlink
+ * allow activation of Generate Forever during generation
+
+
+## 1.3.2
+
+### Bug Fixes:
+ * fix files served out of tmp directory even if they are saved to disk
+ * fix postprocessing overwriting parameters
+
+## 1.3.1
+
+### Features:
+ * revert default cross attention optimization to Doggettx
+
+### Bug Fixes:
+ * fix bug: LoRA don't apply on dropdown list sd_lora
+ * fix png info always added even if setting is not enabled
+ * fix some fields not applying in xyz plot
+ * fix "hires. fix" prompt sharing same labels with txt2img_prompt
+ * fix lora hashes not being added properly to infotex if there is only one lora
+ * fix --use-cpu failing to work properly at startup
+ * make --disable-opt-split-attention command line option work again
+
+## 1.3.0
+
+### Features:
+ * add UI to edit defaults
+ * token merging (via dbolya/tomesd)
+ * settings tab rework: add a lot of additional explanations and links
+ * load extensions' Git metadata in parallel to loading the main program to save a ton of time during startup
+ * update extensions table: show branch, show date in separate column, and show version from tags if available
+ * TAESD - another option for cheap live previews
+ * allow choosing sampler and prompts for second pass of hires fix - hidden by default, enabled in settings
+ * calculate hashes for Lora
+ * add lora hashes to infotext
+ * when pasting infotext, use infotext's lora hashes to find local loras for `<lora:xxx:1>` entries whose hashes match loras the user has
+ * select cross attention optimization from UI
+
+### Minor:
+ * bump Gradio to 3.31.0
+ * bump PyTorch to 2.0.1 for macOS and Linux AMD
+ * allow setting defaults for elements in extensions' tabs
+ * allow selecting file type for live previews
+ * show "Loading..." for extra networks when displaying for the first time
+ * suppress ENSD infotext for samplers that don't use it
+ * clientside optimizations
+ * add options to show/hide hidden files and dirs in extra networks, and to not list models/files in hidden directories
+ * allow whitespace in styles.csv
+ * add option to reorder tabs
+ * move some functionality (swap resolution and set seed to -1) to client
+ * option to specify editor height for img2img
+ * button to copy image resolution into img2img width/height sliders
+ * switch from pyngrok to ngrok-py
+ * lazy-load images in extra networks UI
+ * set "Navigate image viewer with gamepad" option to false by default, by request
+ * change upscalers to download models into user-specified directory (from commandline args) rather than the default models/<...>
+ * allow hiding buttons in ui-config.json
+
+### Extensions:
+ * add /sdapi/v1/script-info api
+ * use Ruff to lint Python code
+ * use ESlint to lint Javascript code
+ * add/modify CFG callbacks for Self-Attention Guidance extension
+ * add command and endpoint for graceful server stopping
+ * add some locals (prompts/seeds/etc) from processing function into the Processing class as fields
+ * rework quoting for infotext items that have commas in them to use JSON (should be backwards compatible except for cases where it didn't work previously)
+ * add /sdapi/v1/refresh-loras api checkpoint post request
+ * tests overhaul
+
+### Bug Fixes:
+ * fix an issue preventing the program from starting if the user specifies a bad Gradio theme
+ * fix broken prompts from file script
+ * fix symlink scanning for extra networks
+ * fix --data-dir ignored when launching via webui-user.bat COMMANDLINE_ARGS
+ * allow web UI to be ran fully offline
+ * fix inability to run with --freeze-settings
+ * fix inability to merge checkpoint without adding metadata
+ * fix extra networks' save preview image not adding infotext for jpeg/webm
+ * remove blinking effect from text in hires fix and scale resolution preview
+ * make links to `http://<...>.git` extensions work in the extension tab
+ * fix bug with webui hanging at startup due to hanging git process
+
+
+## 1.2.1
+
+### Features:
+ * add an option to always refer to LoRA by filenames
+
+### Bug Fixes:
+ * never refer to LoRA by an alias if multiple LoRAs have same alias or the alias is called none
+ * fix upscalers disappearing after the user reloads UI
+ * allow bf16 in safe unpickler (resolves problems with loading some LoRAs)
+ * allow web UI to be ran fully offline
+ * fix localizations not working
+ * fix error for LoRAs: `'LatentDiffusion' object has no attribute 'lora_layer_mapping'`
+
+## 1.2.0
+
+### Features:
+ * do not wait for Stable Diffusion model to load at startup
+ * add filename patterns: `[denoising]`
+ * directory hiding for extra networks: dirs starting with `.` will hide their cards on extra network tabs unless specifically searched for
+ * LoRA: for the `<...>` text in prompt, use name of LoRA that is in the metdata of the file, if present, instead of filename (both can be used to activate LoRA)
+ * LoRA: read infotext params from kohya-ss's extension parameters if they are present and if his extension is not active
+ * LoRA: fix some LoRAs not working (ones that have 3x3 convolution layer)
+ * LoRA: add an option to use old method of applying LoRAs (producing same results as with kohya-ss)
+ * add version to infotext, footer and console output when starting
+ * add links to wiki for filename pattern settings
+ * add extended info for quicksettings setting and use multiselect input instead of a text field
+
+### Minor:
+ * bump Gradio to 3.29.0
+ * bump PyTorch to 2.0.1
+ * `--subpath` option for gradio for use with reverse proxy
+ * Linux/macOS: use existing virtualenv if already active (the VIRTUAL_ENV environment variable)
+ * do not apply localizations if there are none (possible frontend optimization)
+ * add extra `None` option for VAE in XYZ plot
+ * print error to console when batch processing in img2img fails
+ * create HTML for extra network pages only on demand
+ * allow directories starting with `.` to still list their models for LoRA, checkpoints, etc
+ * put infotext options into their own category in settings tab
+ * do not show licenses page when user selects Show all pages in settings
+
+### Extensions:
+ * tooltip localization support
+ * add API method to get LoRA models with prompt
+
+### Bug Fixes:
+ * re-add `/docs` endpoint
+ * fix gamepad navigation
+ * make the lightbox fullscreen image function properly
+ * fix squished thumbnails in extras tab
+ * keep "search" filter for extra networks when user refreshes the tab (previously it showed everthing after you refreshed)
+ * fix webui showing the same image if you configure the generation to always save results into same file
+ * fix bug with upscalers not working properly
+ * fix MPS on PyTorch 2.0.1, Intel Macs
+ * make it so that custom context menu from contextMenu.js only disappears after user's click, ignoring non-user click events
+ * prevent Reload UI button/link from reloading the page when it's not yet ready
+ * fix prompts from file script failing to read contents from a drag/drop file
+
+
+## 1.1.1
+### Bug Fixes:
+ * fix an error that prevents running webui on PyTorch<2.0 without --disable-safe-unpickle
+
+## 1.1.0
+### Features:
+ * switch to PyTorch 2.0.0 (except for AMD GPUs)
+ * visual improvements to custom code scripts
+ * add filename patterns: `[clip_skip]`, `[hasprompt<>]`, `[batch_number]`, `[generation_number]`
+ * add support for saving init images in img2img, and record their hashes in infotext for reproducability
+ * automatically select current word when adjusting weight with ctrl+up/down
+ * add dropdowns for X/Y/Z plot
+ * add setting: Stable Diffusion/Random number generator source: makes it possible to make images generated from a given manual seed consistent across different GPUs
+ * support Gradio's theme API
+ * use TCMalloc on Linux by default; possible fix for memory leaks
+ * add optimization option to remove negative conditioning at low sigma values #9177
+ * embed model merge metadata in .safetensors file
+ * extension settings backup/restore feature #9169
+ * add "resize by" and "resize to" tabs to img2img
+ * add option "keep original size" to textual inversion images preprocess
+ * image viewer scrolling via analog stick
+ * button to restore the progress from session lost / tab reload
+
+### Minor:
+ * bump Gradio to 3.28.1
+ * change "scale to" to sliders in Extras tab
+ * add labels to tool buttons to make it possible to hide them
+ * add tiled inference support for ScuNET
+ * add branch support for extension installation
+ * change Linux installation script to install into current directory rather than `/home/username`
+ * sort textual inversion embeddings by name (case-insensitive)
+ * allow styles.csv to be symlinked or mounted in docker
+ * remove the "do not add watermark to images" option
+ * make selected tab configurable with UI config
+ * make the extra networks UI fixed height and scrollable
+ * add `disable_tls_verify` arg for use with self-signed certs
+
+### Extensions:
+ * add reload callback
+ * add `is_hr_pass` field for processing
+
+### Bug Fixes:
+ * fix broken batch image processing on 'Extras/Batch Process' tab
+ * add "None" option to extra networks dropdowns
+ * fix FileExistsError for CLIP Interrogator
+ * fix /sdapi/v1/txt2img endpoint not working on Linux #9319
+ * fix disappearing live previews and progressbar during slow tasks
+ * fix fullscreen image view not working properly in some cases
+ * prevent alwayson_scripts args param resizing script_arg list when they are inserted in it
+ * fix prompt schedule for second order samplers
+ * fix image mask/composite for weird resolutions #9628
+ * use correct images for previews when using AND (see #9491)
+ * one broken image in img2img batch won't stop all processing
+ * fix image orientation bug in train/preprocess
+ * fix Ngrok recreating tunnels every reload
+ * fix `--realesrgan-models-path` and `--ldsr-models-path` not working
+ * fix `--skip-install` not working
+ * use SAMPLE file format in Outpainting Mk2 & Poorman
+ * do not fail all LoRAs if some have failed to load when making a picture
+
+## 1.0.0
+  * everything

CITATION.cff

@@ -0,0 +1,7 @@
+cff-version: 1.2.0
+message: "If you use this software, please cite it as below."
+authors:
+  - given-names: AUTOMATIC1111
+title: "Stable Diffusion Web UI"
+date-released: 2022-08-22
+url: "https://github.com/AUTOMATIC1111/stable-diffusion-webui"

CODEOWNERS

@@ -0,0 +1 @@
+*       @lllyasviel

LICENSE.txt

@@ -0,0 +1,663 @@
+                    GNU AFFERO GENERAL PUBLIC LICENSE
+                       Version 3, 19 November 2007
+
+                    Copyright (c) 2023 AUTOMATIC1111
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU Affero General Public License is a free, copyleft license for
+software and other kinds of works, specifically designed to ensure
+cooperation with the community in the case of network server software.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+our General Public Licenses are intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  Developers that use our General Public Licenses protect your rights
+with two steps: (1) assert copyright on the software, and (2) offer
+you this License which gives you legal permission to copy, distribute
+and/or modify the software.
+
+  A secondary benefit of defending all users' freedom is that
+improvements made in alternate versions of the program, if they
+receive widespread use, become available for other developers to
+incorporate.  Many developers of free software are heartened and
+encouraged by the resulting cooperation.  However, in the case of
+software used on network servers, this result may fail to come about.
+The GNU General Public License permits making a modified version and
+letting the public access it on a server without ever releasing its
+source code to the public.
+
+  The GNU Affero General Public License is designed specifically to
+ensure that, in such cases, the modified source code becomes available
+to the community.  It requires the operator of a network server to
+provide the source code of the modified version running there to the
+users of that server.  Therefore, public use of a modified version, on
+a publicly accessible server, gives the public access to the source
+code of the modified version.
+
+  An older license, called the Affero General Public License and
+published by Affero, was designed to accomplish similar goals.  This is
+a different license, not a version of the Affero GPL, but Affero has
+released a new version of the Affero GPL which permits relicensing under
+this license.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU Affero General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Remote Network Interaction; Use with the GNU General Public License.
+
+  Notwithstanding any other provision of this License, if you modify the
+Program, your modified version must prominently offer all users
+interacting with it remotely through a computer network (if your version
+supports such interaction) an opportunity to receive the Corresponding
+Source of your version by providing access to the Corresponding Source
+from a network server at no charge, through some standard or customary
+means of facilitating copying of software.  This Corresponding Source
+shall include the Corresponding Source for any work covered by version 3
+of the GNU General Public License that is incorporated pursuant to the
+following paragraph.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the work with which it is combined will remain governed by version
+3 of the GNU General Public License.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU Affero General Public License from time to time.  Such new versions
+will be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU Affero General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU Affero General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU Affero General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU Affero General Public License as published by
+    the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU Affero General Public License for more details.
+
+    You should have received a copy of the GNU Affero General Public License
+    along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If your software can interact with users remotely through a computer
+network, you should also make sure that it provides a way for users to
+get its source.  For example, if your program is a web application, its
+interface could display a "Source" link that leads users to an archive
+of the code.  There are many ways you could offer source, and different
+solutions will be better for different programs; see section 13 for the
+specific requirements.
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU AGPL, see
+<https://www.gnu.org/licenses/>.

README.md

@@ -1,12 +1,710 @@
 ---
-title: Stable Diffusion Webui Forge
-emoji: 🌍
-colorFrom: gray
-colorTo: red
+title: stable-diffusion-webui-forge
+app_file: webui.py
 sdk: gradio
-sdk_version: 4.24.0
-app_file: app.py
-pinned: false
+sdk_version: 3.41.2
 ---
+# Stable Diffusion WebUI Forge
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+Stable Diffusion WebUI Forge is a platform on top of [Stable Diffusion WebUI](https://github.com/AUTOMATIC1111/stable-diffusion-webui) (based on [Gradio](https://www.gradio.app/)) to make development easier, optimize resource management, and speed up inference.
+
+The name "Forge" is inspired from "Minecraft Forge". This project is aimed at becoming SD WebUI's Forge.
+
+Compared to original WebUI (for SDXL inference at 1024px), you can expect the below speed-ups:
+
+1. If you use common GPU like 8GB vram, you can expect to get about **30~45% speed up** in inference speed (it/s), the GPU memory peak (in task manager) will drop about 700MB to 1.3GB, the maximum diffusion resolution (that will not OOM) will increase about 2x to 3x, and the maximum diffusion batch size (that will not OOM) will increase about 4x to 6x.
+
+2. If you use less powerful GPU like 6GB vram, you can expect to get about **60~75% speed up** in inference speed (it/s), the GPU memory peak (in task manager) will drop about 800MB to 1.5GB, the maximum diffusion resolution (that will not OOM) will increase about 3x, and the maximum diffusion batch size (that will not OOM) will increase about 4x.
+
+3. If you use powerful GPU like 4090 with 24GB vram, you can expect to get about **3~6% speed up** in inference speed (it/s), the GPU memory peak (in task manager) will drop about 1GB to 1.4GB, the maximum diffusion resolution (that will not OOM) will increase about 1.6x, and the maximum diffusion batch size (that will not OOM) will increase about 2x.
+
+4. If you use ControlNet for SDXL, the maximum ControlNet count (that will not OOM) will increase about 2x, the speed with SDXL+ControlNet will **speed up about 30~45%**.
+
+Another very important change that Forge brings is **Unet Patcher**. Using Unet Patcher, methods like Self-Attention Guidance, Kohya High Res Fix, FreeU, StyleAlign, Hypertile can all be implemented in about 100 lines of codes. 
+
+Thanks to Unet Patcher, many new things are possible now and supported in Forge, including SVD, Z123, masked Ip-adapter, masked controlnet, photomaker, etc.
+
+**No need to monkeypatch UNet and conflict other extensions anymore!**
+
+Forge also adds a few samplers, including but not limited to DDPM, DDPM Karras, DPM++ 2M Turbo, DPM++ 2M SDE Turbo, LCM Karras, Euler A Turbo, etc. (LCM is already in original webui since 1.7.0).
+
+Finally, Forge promise that we will only do our jobs. Forge will never add unnecessary opinioned changes to the user interface. You are still using 100% Automatic1111 WebUI.
+
+# Installing Forge
+
+If you are proficient in Git and you want to install Forge as another branch of SD-WebUI, please see [here](https://github.com/continue-revolution/sd-webui-animatediff/blob/forge/master/docs/how-to-use.md#you-have-a1111-and-you-know-git). In this way, you can reuse all SD checkpoints and all extensions you installed previously in your OG SD-WebUI, but you should know what you are doing.
+
+If you know what you are doing, you can install Forge using same method as SD-WebUI. (Install Git, Python, Git Clone the forge repo `https://github.com/lllyasviel/stable-diffusion-webui-forge.git` and then run webui-user.bat).
+
+**Or you can just use this one-click installation package (with git and python included).**
+
+[>>> Click Here to Download One-Click Package<<<](https://github.com/lllyasviel/stable-diffusion-webui-forge/releases/download/latest/webui_forge_cu121_torch21.7z)
+
+After you download, you uncompress, use `update.bat` to update, and use `run.bat` to run.
+
+Note that running `update.bat` is important, otherwise you may be using a previous version with potential bugs unfixed.
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/c49bd60d-82bd-4086-9859-88d472582b94)
+
+# Screenshots of Comparison
+
+I tested with several devices, and this is a typical result from 8GB VRAM (3070ti laptop) with SDXL.
+
+**This is original WebUI:**
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/16893937-9ed9-4f8e-b960-70cd5d1e288f)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/7bbc16fe-64ef-49e2-a595-d91bb658bd94)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/de1747fd-47bc-482d-a5c6-0728dd475943)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/96e5e171-2d74-41ba-9dcc-11bf68be7e16)
+
+(average about 7.4GB/8GB, peak at about 7.9GB/8GB)
+
+**This is WebUI Forge:**
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/ca5e05ed-bd86-4ced-8662-f41034648e8c)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/3629ee36-4a99-4d9b-b371-12efb260a283)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/6d13ebb7-c30d-4aa8-9242-c0b5a1af8c95)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/c4f723c3-6ea7-4539-980b-0708ed2a69aa)
+
+(average and peak are all 6.3GB/8GB)
+
+You can see that Forge does not change WebUI results. Installing Forge is not a seed breaking change. 
+
+Forge can perfectly keep WebUI unchanged even for most complicated prompts like `fantasy landscape with a [mountain:lake:0.25] and [an oak:a christmas tree:0.75][ in foreground::0.6][ in background:0.25] [shoddy:masterful:0.5]`.
+
+All your previous works still work in Forge!
+
+# Forge Backend
+
+Forge backend removes all WebUI's codes related to resource management and reworked everything. All previous CMD flags like `medvram, lowvram, medvram-sdxl, precision full, no half, no half vae, attention_xxx, upcast unet`, ... are all **REMOVED**. Adding these flags will not cause error but they will not do anything now. **We highly encourage Forge users to remove all cmd flags and let Forge to decide how to load models.**
+
+Without any cmd flag, Forge can run SDXL with 4GB vram and SD1.5 with 2GB vram.
+
+**Some flags that you may still pay attention to:** 
+
+1. `--always-offload-from-vram` (This flag will make things **slower** but less risky). This option will let Forge always unload models from VRAM. This can be useful if you use multiple software together and want Forge to use less VRAM and give some VRAM to other software, or when you are using some old extensions that will compete vram with Forge, or (very rarely) when you get OOM.
+
+2. `--cuda-malloc` (This flag will make things **faster** but more risky). This will ask pytorch to use *cudaMallocAsync* for tensor malloc. On some profilers I can observe performance gain at millisecond level, but the real speed up on most my devices are often unnoticed (about or less than 0.1 second per image). This cannot be set as default because many users reported issues that the async malloc will crash the program. Users need to enable this cmd flag at their own risk.
+
+3. `--cuda-stream` (This flag will make things **faster** but more risky). This will use pytorch CUDA streams (a special type of thread on GPU) to move models and compute tensors simultaneously. This can almost eliminate all model moving time, and speed up SDXL on 30XX/40XX devices with small VRAM (eg, RTX 4050 6GB, RTX 3060 Laptop 6GB, etc) by about 15\% to 25\%. However, this unfortunately cannot be set as default because I observe higher possibility of pure black images (Nan outputs) on 2060, and higher chance of OOM on 1080 and 2060. When the resolution is large, there is a chance that the computation time of one single attention layer is longer than the time for moving entire model to GPU. When that happens, the next attention layer will OOM since the GPU is filled with the entire model, and no remaining space is available for computing another attention layer. Most overhead detecting methods are not robust enough to be reliable on old devices (in my tests). Users need to enable this cmd flag at their own risk.
+
+4. `--pin-shared-memory` (This flag will make things **faster** but more risky). Effective only when used together with `--cuda-stream`. This will offload modules to Shared GPU Memory instead of system RAM when offloading models. On some 30XX/40XX devices with small VRAM (eg, RTX 4050 6GB, RTX 3060 Laptop 6GB, etc), I can observe significant (at least 20\%) speed-up for SDXL. However, this unfortunately cannot be set as default because the OOM of Shared GPU Memory is a much more severe problem than common GPU memory OOM. Pytorch does not provide any robust method to unload or detect Shared GPU Memory. Once the Shared GPU Memory OOM, the entire program will crash (observed with SDXL on GTX 1060/1050/1066), and there is no dynamic method to prevent or recover from the crash. Users need to enable this cmd flag at their own risk.
+
+If you really want to play with cmd flags, you can additionally control the GPU with:
+
+(extreme VRAM cases)
+
+    --always-gpu
+    --always-cpu
+
+(rare attention cases)
+
+    --attention-split
+    --attention-quad
+    --attention-pytorch
+    --disable-xformers
+    --disable-attention-upcast
+
+(float point type)
+
+    --all-in-fp32
+    --all-in-fp16
+    --unet-in-bf16
+    --unet-in-fp16
+    --unet-in-fp8-e4m3fn
+    --unet-in-fp8-e5m2
+    --vae-in-fp16
+    --vae-in-fp32
+    --vae-in-bf16
+    --clip-in-fp8-e4m3fn
+    --clip-in-fp8-e5m2
+    --clip-in-fp16
+    --clip-in-fp32
+
+(rare platforms)
+
+    --directml
+    --disable-ipex-hijack
+    --pytorch-deterministic
+
+Again, Forge do not recommend users to use any cmd flags unless you are very sure that you really need these.
+
+# UNet Patcher
+
+Note that [Forge does not use any other software as backend](https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/169). The full name of the backend is `Stable Diffusion WebUI with Forge backend`, or for simplicity, the `Forge backend`. The API and python symbols are made similar to previous software only for reducing the learning cost of developers.
+
+Now developing an extension is super simple. We finally have a patchable UNet.
+
+Below is using one single file with 80 lines of codes to support FreeU:
+
+`extensions-builtin/sd_forge_freeu/scripts/forge_freeu.py`
+
+```python
+import torch
+import gradio as gr
+from modules import scripts
+
+
+def Fourier_filter(x, threshold, scale):
+    x_freq = torch.fft.fftn(x.float(), dim=(-2, -1))
+    x_freq = torch.fft.fftshift(x_freq, dim=(-2, -1))
+    B, C, H, W = x_freq.shape
+    mask = torch.ones((B, C, H, W), device=x.device)
+    crow, ccol = H // 2, W //2
+    mask[..., crow - threshold:crow + threshold, ccol - threshold:ccol + threshold] = scale
+    x_freq = x_freq * mask
+    x_freq = torch.fft.ifftshift(x_freq, dim=(-2, -1))
+    x_filtered = torch.fft.ifftn(x_freq, dim=(-2, -1)).real
+    return x_filtered.to(x.dtype)
+
+
+def set_freeu_v2_patch(model, b1, b2, s1, s2):
+    model_channels = model.model.model_config.unet_config["model_channels"]
+    scale_dict = {model_channels * 4: (b1, s1), model_channels * 2: (b2, s2)}
+
+    def output_block_patch(h, hsp, *args, **kwargs):
+        scale = scale_dict.get(h.shape[1], None)
+        if scale is not None:
+            hidden_mean = h.mean(1).unsqueeze(1)
+            B = hidden_mean.shape[0]
+            hidden_max, _ = torch.max(hidden_mean.view(B, -1), dim=-1, keepdim=True)
+            hidden_min, _ = torch.min(hidden_mean.view(B, -1), dim=-1, keepdim=True)
+            hidden_mean = (hidden_mean - hidden_min.unsqueeze(2).unsqueeze(3)) / \
+                          (hidden_max - hidden_min).unsqueeze(2).unsqueeze(3)
+            h[:, :h.shape[1] // 2] = h[:, :h.shape[1] // 2] * ((scale[0] - 1) * hidden_mean + 1)
+            hsp = Fourier_filter(hsp, threshold=1, scale=scale[1])
+        return h, hsp
+
+    m = model.clone()
+    m.set_model_output_block_patch(output_block_patch)
+    return m
+
+
+class FreeUForForge(scripts.Script):
+    def title(self):
+        return "FreeU Integrated"
+
+    def show(self, is_img2img):
+        # make this extension visible in both txt2img and img2img tab.
+        return scripts.AlwaysVisible
+
+    def ui(self, *args, **kwargs):
+        with gr.Accordion(open=False, label=self.title()):
+            freeu_enabled = gr.Checkbox(label='Enabled', value=False)
+            freeu_b1 = gr.Slider(label='B1', minimum=0, maximum=2, step=0.01, value=1.01)
+            freeu_b2 = gr.Slider(label='B2', minimum=0, maximum=2, step=0.01, value=1.02)
+            freeu_s1 = gr.Slider(label='S1', minimum=0, maximum=4, step=0.01, value=0.99)
+            freeu_s2 = gr.Slider(label='S2', minimum=0, maximum=4, step=0.01, value=0.95)
+
+        return freeu_enabled, freeu_b1, freeu_b2, freeu_s1, freeu_s2
+
+    def process_before_every_sampling(self, p, *script_args, **kwargs):
+        # This will be called before every sampling.
+        # If you use highres fix, this will be called twice.
+        
+        freeu_enabled, freeu_b1, freeu_b2, freeu_s1, freeu_s2 = script_args
+
+        if not freeu_enabled:
+            return
+
+        unet = p.sd_model.forge_objects.unet
+
+        unet = set_freeu_v2_patch(unet, freeu_b1, freeu_b2, freeu_s1, freeu_s2)
+
+        p.sd_model.forge_objects.unet = unet
+
+        # Below codes will add some logs to the texts below the image outputs on UI.
+        # The extra_generation_params does not influence results.
+        p.extra_generation_params.update(dict(
+            freeu_enabled=freeu_enabled,
+            freeu_b1=freeu_b1,
+            freeu_b2=freeu_b2,
+            freeu_s1=freeu_s1,
+            freeu_s2=freeu_s2,
+        ))
+
+        return
+```
+
+It looks like this:
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/277bac6e-5ea7-4bff-b71a-e55a60cfc03c)
+
+Similar components like HyperTile, KohyaHighResFix, SAG, can all be implemented within 100 lines of codes (see also the codes).
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/06472b03-b833-4816-ab47-70712ac024d3)
+
+ControlNets can finally be called by different extensions.
+
+Implementing Stable Video Diffusion and Zero123 are also super simple now (see also the codes). 
+
+*Stable Video Diffusion:*
+
+`extensions-builtin/sd_forge_svd/scripts/forge_svd.py`
+
+```python
+import torch
+import gradio as gr
+import os
+import pathlib
+
+from modules import script_callbacks
+from modules.paths import models_path
+from modules.ui_common import ToolButton, refresh_symbol
+from modules import shared
+
+from modules_forge.forge_util import numpy_to_pytorch, pytorch_to_numpy
+from ldm_patched.modules.sd import load_checkpoint_guess_config
+from ldm_patched.contrib.external_video_model import VideoLinearCFGGuidance, SVD_img2vid_Conditioning
+from ldm_patched.contrib.external import KSampler, VAEDecode
+
+
+opVideoLinearCFGGuidance = VideoLinearCFGGuidance()
+opSVD_img2vid_Conditioning = SVD_img2vid_Conditioning()
+opKSampler = KSampler()
+opVAEDecode = VAEDecode()
+
+svd_root = os.path.join(models_path, 'svd')
+os.makedirs(svd_root, exist_ok=True)
+svd_filenames = []
+
+
+def update_svd_filenames():
+    global svd_filenames
+    svd_filenames = [
+        pathlib.Path(x).name for x in
+        shared.walk_files(svd_root, allowed_extensions=[".pt", ".ckpt", ".safetensors"])
+    ]
+    return svd_filenames
+
+
+@torch.inference_mode()
+@torch.no_grad()
+def predict(filename, width, height, video_frames, motion_bucket_id, fps, augmentation_level,
+            sampling_seed, sampling_steps, sampling_cfg, sampling_sampler_name, sampling_scheduler,
+            sampling_denoise, guidance_min_cfg, input_image):
+    filename = os.path.join(svd_root, filename)
+    model_raw, _, vae, clip_vision = \
+        load_checkpoint_guess_config(filename, output_vae=True, output_clip=False, output_clipvision=True)
+    model = opVideoLinearCFGGuidance.patch(model_raw, guidance_min_cfg)[0]
+    init_image = numpy_to_pytorch(input_image)
+    positive, negative, latent_image = opSVD_img2vid_Conditioning.encode(
+        clip_vision, init_image, vae, width, height, video_frames, motion_bucket_id, fps, augmentation_level)
+    output_latent = opKSampler.sample(model, sampling_seed, sampling_steps, sampling_cfg,
+                                      sampling_sampler_name, sampling_scheduler, positive,
+                                      negative, latent_image, sampling_denoise)[0]
+    output_pixels = opVAEDecode.decode(vae, output_latent)[0]
+    outputs = pytorch_to_numpy(output_pixels)
+    return outputs
+
+
+def on_ui_tabs():
+    with gr.Blocks() as svd_block:
+        with gr.Row():
+            with gr.Column():
+                input_image = gr.Image(label='Input Image', source='upload', type='numpy', height=400)
+
+                with gr.Row():
+                    filename = gr.Dropdown(label="SVD Checkpoint Filename",
+                                           choices=svd_filenames,
+                                           value=svd_filenames[0] if len(svd_filenames) > 0 else None)
+                    refresh_button = ToolButton(value=refresh_symbol, tooltip="Refresh")
+                    refresh_button.click(
+                        fn=lambda: gr.update(choices=update_svd_filenames),
+                        inputs=[], outputs=filename)
+
+                width = gr.Slider(label='Width', minimum=16, maximum=8192, step=8, value=1024)
+                height = gr.Slider(label='Height', minimum=16, maximum=8192, step=8, value=576)
+                video_frames = gr.Slider(label='Video Frames', minimum=1, maximum=4096, step=1, value=14)
+                motion_bucket_id = gr.Slider(label='Motion Bucket Id', minimum=1, maximum=1023, step=1, value=127)
+                fps = gr.Slider(label='Fps', minimum=1, maximum=1024, step=1, value=6)
+                augmentation_level = gr.Slider(label='Augmentation Level', minimum=0.0, maximum=10.0, step=0.01,
+                                               value=0.0)
+                sampling_steps = gr.Slider(label='Sampling Steps', minimum=1, maximum=200, step=1, value=20)
+                sampling_cfg = gr.Slider(label='CFG Scale', minimum=0.0, maximum=50.0, step=0.1, value=2.5)
+                sampling_denoise = gr.Slider(label='Sampling Denoise', minimum=0.0, maximum=1.0, step=0.01, value=1.0)
+                guidance_min_cfg = gr.Slider(label='Guidance Min Cfg', minimum=0.0, maximum=100.0, step=0.5, value=1.0)
+                sampling_sampler_name = gr.Radio(label='Sampler Name',
+                                                 choices=['euler', 'euler_ancestral', 'heun', 'heunpp2', 'dpm_2',
+                                                          'dpm_2_ancestral', 'lms', 'dpm_fast', 'dpm_adaptive',
+                                                          'dpmpp_2s_ancestral', 'dpmpp_sde', 'dpmpp_sde_gpu',
+                                                          'dpmpp_2m', 'dpmpp_2m_sde', 'dpmpp_2m_sde_gpu',
+                                                          'dpmpp_3m_sde', 'dpmpp_3m_sde_gpu', 'ddpm', 'lcm', 'ddim',
+                                                          'uni_pc', 'uni_pc_bh2'], value='euler')
+                sampling_scheduler = gr.Radio(label='Scheduler',
+                                              choices=['normal', 'karras', 'exponential', 'sgm_uniform', 'simple',
+                                                       'ddim_uniform'], value='karras')
+                sampling_seed = gr.Number(label='Seed', value=12345, precision=0)
+
+                generate_button = gr.Button(value="Generate")
+
+                ctrls = [filename, width, height, video_frames, motion_bucket_id, fps, augmentation_level,
+                         sampling_seed, sampling_steps, sampling_cfg, sampling_sampler_name, sampling_scheduler,
+                         sampling_denoise, guidance_min_cfg, input_image]
+
+            with gr.Column():
+                output_gallery = gr.Gallery(label='Gallery', show_label=False, object_fit='contain',
+                                            visible=True, height=1024, columns=4)
+
+        generate_button.click(predict, inputs=ctrls, outputs=[output_gallery])
+    return [(svd_block, "SVD", "svd")]
+
+
+update_svd_filenames()
+script_callbacks.on_ui_tabs(on_ui_tabs)
+```
+
+Note that although the above codes look like independent codes, they actually will automatically offload/unload any other models. For example, below is me opening webui, load SDXL, generated an image, then go to SVD, then generated image frames. You can see that the GPU memory is perfectly managed and the SDXL is moved to RAM then SVD is moved to GPU. 
+
+Note that this management is fully automatic. This makes writing extensions super simple.
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/de1a2d05-344a-44d7-bab8-9ecc0a58a8d3)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/14bcefcf-599f-42c3-bce9-3fd5e428dd91)
+
+Similarly, Zero123:
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/7685019c-7239-47fb-9cb5-2b7b33943285)
+
+### Write a simple ControlNet:
+
+Below is a simple extension to have a completely independent pass of ControlNet that never conflicts any other extensions:
+
+`extensions-builtin/sd_forge_controlnet_example/scripts/sd_forge_controlnet_example.py`
+
+Note that this extension is hidden because it is only for developers. To see it in UI, use `--show-controlnet-example`.
+
+The memory optimization in this example is fully automatic. You do not need to care about memory and inference speed, but you may want to cache objects if you wish.
+
+```python
+# Use --show-controlnet-example to see this extension.
+
+import cv2
+import gradio as gr
+import torch
+
+from modules import scripts
+from modules.shared_cmd_options import cmd_opts
+from modules_forge.shared import supported_preprocessors
+from modules.modelloader import load_file_from_url
+from ldm_patched.modules.controlnet import load_controlnet
+from modules_forge.controlnet import apply_controlnet_advanced
+from modules_forge.forge_util import numpy_to_pytorch
+from modules_forge.shared import controlnet_dir
+
+
+class ControlNetExampleForge(scripts.Script):
+    model = None
+
+    def title(self):
+        return "ControlNet Example for Developers"
+
+    def show(self, is_img2img):
+        # make this extension visible in both txt2img and img2img tab.
+        return scripts.AlwaysVisible
+
+    def ui(self, *args, **kwargs):
+        with gr.Accordion(open=False, label=self.title()):
+            gr.HTML('This is an example controlnet extension for developers.')
+            gr.HTML('You see this extension because you used --show-controlnet-example')
+            input_image = gr.Image(source='upload', type='numpy')
+            funny_slider = gr.Slider(label='This slider does nothing. It just shows you how to transfer parameters.',
+                                     minimum=0.0, maximum=1.0, value=0.5)
+
+        return input_image, funny_slider
+
+    def process(self, p, *script_args, **kwargs):
+        input_image, funny_slider = script_args
+
+        # This slider does nothing. It just shows you how to transfer parameters.
+        del funny_slider
+
+        if input_image is None:
+            return
+
+        # controlnet_canny_path = load_file_from_url(
+        #     url='https://huggingface.co/lllyasviel/sd_control_collection/resolve/main/sai_xl_canny_256lora.safetensors',
+        #     model_dir=model_dir,
+        #     file_name='sai_xl_canny_256lora.safetensors'
+        # )
+        controlnet_canny_path = load_file_from_url(
+            url='https://huggingface.co/lllyasviel/fav_models/resolve/main/fav/control_v11p_sd15_canny_fp16.safetensors',
+            model_dir=controlnet_dir,
+            file_name='control_v11p_sd15_canny_fp16.safetensors'
+        )
+        print('The model [control_v11p_sd15_canny_fp16.safetensors] download finished.')
+
+        self.model = load_controlnet(controlnet_canny_path)
+        print('Controlnet loaded.')
+
+        return
+
+    def process_before_every_sampling(self, p, *script_args, **kwargs):
+        # This will be called before every sampling.
+        # If you use highres fix, this will be called twice.
+
+        input_image, funny_slider = script_args
+
+        if input_image is None or self.model is None:
+            return
+
+        B, C, H, W = kwargs['noise'].shape  # latent_shape
+        height = H * 8
+        width = W * 8
+        batch_size = p.batch_size
+
+        preprocessor = supported_preprocessors['canny']
+
+        # detect control at certain resolution
+        control_image = preprocessor(
+            input_image, resolution=512, slider_1=100, slider_2=200, slider_3=None)
+
+        # here we just use nearest neighbour to align input shape.
+        # You may want crop and resize, or crop and fill, or others.
+        control_image = cv2.resize(
+            control_image, (width, height), interpolation=cv2.INTER_NEAREST)
+
+        # Output preprocessor result. Now called every sampling. Cache in your own way.
+        p.extra_result_images.append(control_image)
+
+        print('Preprocessor Canny finished.')
+
+        control_image_bchw = numpy_to_pytorch(control_image).movedim(-1, 1)
+
+        unet = p.sd_model.forge_objects.unet
+
+        # Unet has input, middle, output blocks, and we can give different weights
+        # to each layers in all blocks.
+        # Below is an example for stronger control in middle block.
+        # This is helpful for some high-res fix passes. (p.is_hr_pass)
+        positive_advanced_weighting = {
+            'input': [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2],
+            'middle': [1.0],
+            'output': [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2]
+        }
+        negative_advanced_weighting = {
+            'input': [0.15, 0.25, 0.35, 0.45, 0.55, 0.65, 0.75, 0.85, 0.95, 1.05, 1.15, 1.25],
+            'middle': [1.05],
+            'output': [0.15, 0.25, 0.35, 0.45, 0.55, 0.65, 0.75, 0.85, 0.95, 1.05, 1.15, 1.25]
+        }
+
+        # The advanced_frame_weighting is a weight applied to each image in a batch.
+        # The length of this list must be same with batch size
+        # For example, if batch size is 5, the below list is [0.2, 0.4, 0.6, 0.8, 1.0]
+        # If you view the 5 images as 5 frames in a video, this will lead to
+        # progressively stronger control over time.
+        advanced_frame_weighting = [float(i + 1) / float(batch_size) for i in range(batch_size)]
+
+        # The advanced_sigma_weighting allows you to dynamically compute control
+        # weights given diffusion timestep (sigma).
+        # For example below code can softly make beginning steps stronger than ending steps.
+        sigma_max = unet.model.model_sampling.sigma_max
+        sigma_min = unet.model.model_sampling.sigma_min
+        advanced_sigma_weighting = lambda s: (s - sigma_min) / (sigma_max - sigma_min)
+
+        # You can even input a tensor to mask all control injections
+        # The mask will be automatically resized during inference in UNet.
+        # The size should be B 1 H W and the H and W are not important
+        # because they will be resized automatically
+        advanced_mask_weighting = torch.ones(size=(1, 1, 512, 512))
+
+        # But in this simple example we do not use them
+        positive_advanced_weighting = None
+        negative_advanced_weighting = None
+        advanced_frame_weighting = None
+        advanced_sigma_weighting = None
+        advanced_mask_weighting = None
+
+        unet = apply_controlnet_advanced(unet=unet, controlnet=self.model, image_bchw=control_image_bchw,
+                                         strength=0.6, start_percent=0.0, end_percent=0.8,
+                                         positive_advanced_weighting=positive_advanced_weighting,
+                                         negative_advanced_weighting=negative_advanced_weighting,
+                                         advanced_frame_weighting=advanced_frame_weighting,
+                                         advanced_sigma_weighting=advanced_sigma_weighting,
+                                         advanced_mask_weighting=advanced_mask_weighting)
+
+        p.sd_model.forge_objects.unet = unet
+
+        # Below codes will add some logs to the texts below the image outputs on UI.
+        # The extra_generation_params does not influence results.
+        p.extra_generation_params.update(dict(
+            controlnet_info='You should see these texts below output images!',
+        ))
+
+        return
+
+
+# Use --show-controlnet-example to see this extension.
+if not cmd_opts.show_controlnet_example:
+    del ControlNetExampleForge
+
+```
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/822fa2fc-c9f4-4f58-8669-4b6680b91063)
+
+
+### Add a preprocessor
+
+Below is the full codes to add a normalbae preprocessor with perfect memory managements.
+
+You can use arbitrary independent extensions to add a preprocessor.
+
+Your preprocessor will be read by all other extensions using `modules_forge.shared.preprocessors`
+
+Below codes are in `extensions-builtin\forge_preprocessor_normalbae\scripts\preprocessor_normalbae.py`
+
+```python
+from modules_forge.supported_preprocessor import Preprocessor, PreprocessorParameter
+from modules_forge.shared import preprocessor_dir, add_supported_preprocessor
+from modules_forge.forge_util import resize_image_with_pad
+from modules.modelloader import load_file_from_url
+
+import types
+import torch
+import numpy as np
+
+from einops import rearrange
+from annotator.normalbae.models.NNET import NNET
+from annotator.normalbae import load_checkpoint
+from torchvision import transforms
+
+
+class PreprocessorNormalBae(Preprocessor):
+    def __init__(self):
+        super().__init__()
+        self.name = 'normalbae'
+        self.tags = ['NormalMap']
+        self.model_filename_filters = ['normal']
+        self.slider_resolution = PreprocessorParameter(
+            label='Resolution', minimum=128, maximum=2048, value=512, step=8, visible=True)
+        self.slider_1 = PreprocessorParameter(visible=False)
+        self.slider_2 = PreprocessorParameter(visible=False)
+        self.slider_3 = PreprocessorParameter(visible=False)
+        self.show_control_mode = True
+        self.do_not_need_model = False
+        self.sorting_priority = 100  # higher goes to top in the list
+
+    def load_model(self):
+        if self.model_patcher is not None:
+            return
+
+        model_path = load_file_from_url(
+            "https://huggingface.co/lllyasviel/Annotators/resolve/main/scannet.pt",
+            model_dir=preprocessor_dir)
+
+        args = types.SimpleNamespace()
+        args.mode = 'client'
+        args.architecture = 'BN'
+        args.pretrained = 'scannet'
+        args.sampling_ratio = 0.4
+        args.importance_ratio = 0.7
+        model = NNET(args)
+        model = load_checkpoint(model_path, model)
+        self.norm = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+
+        self.model_patcher = self.setup_model_patcher(model)
+
+    def __call__(self, input_image, resolution, slider_1=None, slider_2=None, slider_3=None, **kwargs):
+        input_image, remove_pad = resize_image_with_pad(input_image, resolution)
+
+        self.load_model()
+
+        self.move_all_model_patchers_to_gpu()
+
+        assert input_image.ndim == 3
+        image_normal = input_image
+
+        with torch.no_grad():
+            image_normal = self.send_tensor_to_model_device(torch.from_numpy(image_normal))
+            image_normal = image_normal / 255.0
+            image_normal = rearrange(image_normal, 'h w c -> 1 c h w')
+            image_normal = self.norm(image_normal)
+
+            normal = self.model_patcher.model(image_normal)
+            normal = normal[0][-1][:, :3]
+            normal = ((normal + 1) * 0.5).clip(0, 1)
+
+            normal = rearrange(normal[0], 'c h w -> h w c').cpu().numpy()
+            normal_image = (normal * 255.0).clip(0, 255).astype(np.uint8)
+
+        return remove_pad(normal_image)
+
+
+add_supported_preprocessor(PreprocessorNormalBae())
+
+```
+
+# New features (that are not available in original WebUI)
+
+Thanks to Unet Patcher, many new things are possible now and supported in Forge, including SVD, Z123, masked Ip-adapter, masked controlnet, photomaker, etc.
+
+Masked Ip-Adapter
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/d26630f9-922d-4483-8bf9-f364dca5fd50)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/03580ef7-235c-4b03-9ca6-a27677a5a175)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/d9ed4a01-70d4-45b4-a6a7-2f765f158fae)
+
+Masked ControlNet
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/872d4785-60e4-4431-85c7-665c781dddaa)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/335a3b33-1ef8-46ff-a462-9f1b4f2c49fc)
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/b3684a15-8895-414e-8188-487269dfcada)
+
+PhotoMaker
+
+(Note that photomaker is a special control that need you to add the trigger word "photomaker". Your prompt should be like "a photo of photomaker")
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/07b0b626-05b5-473b-9d69-3657624d59be)
+
+Marigold Depth
+
+![image](https://github.com/lllyasviel/stable-diffusion-webui-forge/assets/19834515/bdf54148-892d-410d-8ed9-70b4b121b6e7)
+
+# New Samplers (that are not in origin)
+
+    DDPM
+    DDPM Karras
+    DPM++ 2M Turbo
+    DPM++ 2M SDE Turbo
+    LCM Karras
+    Euler A Turbo
+
+# About Extensions
+
+ControlNet and TiledVAE are integrated, and you should uninstall these two extensions:
+
+    sd-webui-controlnet
+    multidiffusion-upscaler-for-automatic1111
+
+Note that **AnimateDiff** is under construction by [continue-revolution](https://github.com/continue-revolution) at [sd-webui-animatediff forge/master branch](https://github.com/continue-revolution/sd-webui-animatediff/tree/forge/master) and [sd-forge-animatediff](https://github.com/continue-revolution/sd-forge-animatediff) (they are in sync). (continue-revolution original words: prompt travel, inf t2v, controlnet v2v have been proven to work well; motion lora, i2i batch still under construction and may be finished in a week")
+
+Other extensions should work without problems, like:
+
+    canvas-zoom
+    translations/localizations
+    Dynamic Prompts
+    Adetailer
+    Ultimate SD Upscale
+    Reactor
+
+However, if newer extensions use Forge, their codes can be much shorter. 
+
+Usually if an old extension rework using Forge's unet patcher, 80% codes can be removed, especially when they need to call controlnet.
+
+# Contribution
+
+Forge uses a bot to get commits and codes from https://github.com/AUTOMATIC1111/stable-diffusion-webui/tree/dev every afternoon (if merge is automatically successful by a git bot, or by my compiler, or by my ChatGPT bot) or mid-night (if my compiler and my ChatGPT bot both failed to merge and I review it manually).
+
+All PRs that can be implemented in https://github.com/AUTOMATIC1111/stable-diffusion-webui/tree/dev should submit PRs there.
+
+Feel free to submit PRs related to the functionality of Forge here.

configs/alt-diffusion-inference.yaml

@@ -0,0 +1,72 @@
+model:
+  base_learning_rate: 1.0e-04
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: "jpg"
+    cond_stage_key: "txt"
+    image_size: 64
+    channels: 4
+    cond_stage_trainable: false   # Note: different from the one we trained before
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+
+    scheduler_config: # 10000 warmup steps
+      target: ldm.lr_scheduler.LambdaLinearScheduler
+      params:
+        warm_up_steps: [ 10000 ]
+        cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases
+        f_start: [ 1.e-6 ]
+        f_max: [ 1. ]
+        f_min: [ 1. ]
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32 # unused
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_heads: 8
+        use_spatial_transformer: True
+        transformer_depth: 1
+        context_dim: 768
+        use_checkpoint: True
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: modules.xlmr.BertSeriesModelWithTransformation
+      params:
+        name: "XLMR-Large"

configs/alt-diffusion-m18-inference.yaml

@@ -0,0 +1,73 @@
+model:
+  base_learning_rate: 1.0e-04
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: "jpg"
+    cond_stage_key: "txt"
+    image_size: 64
+    channels: 4
+    cond_stage_trainable: false   # Note: different from the one we trained before
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+
+    scheduler_config: # 10000 warmup steps
+      target: ldm.lr_scheduler.LambdaLinearScheduler
+      params:
+        warm_up_steps: [ 10000 ]
+        cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases
+        f_start: [ 1.e-6 ]
+        f_max: [ 1. ]
+        f_min: [ 1. ]
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32 # unused
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_head_channels: 64
+        use_spatial_transformer: True
+        use_linear_in_transformer: True
+        transformer_depth: 1
+        context_dim: 1024
+        use_checkpoint: True
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: modules.xlmr_m18.BertSeriesModelWithTransformation
+      params:
+        name: "XLMR-Large"

configs/instruct-pix2pix.yaml

@@ -0,0 +1,98 @@
+# File modified by authors of InstructPix2Pix from original (https://github.com/CompVis/stable-diffusion).
+# See more details in LICENSE.
+
+model:
+  base_learning_rate: 1.0e-04
+  target: modules.models.diffusion.ddpm_edit.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: edited
+    cond_stage_key: edit
+    # image_size: 64
+    # image_size: 32
+    image_size: 16
+    channels: 4
+    cond_stage_trainable: false   # Note: different from the one we trained before
+    conditioning_key: hybrid
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: false
+
+    scheduler_config: # 10000 warmup steps
+      target: ldm.lr_scheduler.LambdaLinearScheduler
+      params:
+        warm_up_steps: [ 0 ]
+        cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases
+        f_start: [ 1.e-6 ]
+        f_max: [ 1. ]
+        f_min: [ 1. ]
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32 # unused
+        in_channels: 8
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_heads: 8
+        use_spatial_transformer: True
+        transformer_depth: 1
+        context_dim: 768
+        use_checkpoint: True
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder
+
+data:
+  target: main.DataModuleFromConfig
+  params:
+    batch_size: 128
+    num_workers: 1
+    wrap: false
+    validation:
+      target: edit_dataset.EditDataset
+      params:
+        path: data/clip-filtered-dataset
+        cache_dir:  data/
+        cache_name: data_10k
+        split: val
+        min_text_sim: 0.2
+        min_image_sim: 0.75
+        min_direction_sim: 0.2
+        max_samples_per_prompt: 1
+        min_resize_res: 512
+        max_resize_res: 512
+        crop_res: 512
+        output_as_edit: False
+        real_input: True

configs/sd_xl_inpaint.yaml

@@ -0,0 +1,98 @@
+model:
+  target: sgm.models.diffusion.DiffusionEngine
+  params:
+    scale_factor: 0.13025
+    disable_first_stage_autocast: True
+
+    denoiser_config:
+      target: sgm.modules.diffusionmodules.denoiser.DiscreteDenoiser
+      params:
+        num_idx: 1000
+
+        weighting_config:
+          target: sgm.modules.diffusionmodules.denoiser_weighting.EpsWeighting
+        scaling_config:
+          target: sgm.modules.diffusionmodules.denoiser_scaling.EpsScaling
+        discretization_config:
+          target: sgm.modules.diffusionmodules.discretizer.LegacyDDPMDiscretization
+
+    network_config:
+      target: sgm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        adm_in_channels: 2816
+        num_classes: sequential
+        use_checkpoint: True
+        in_channels: 9
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [4, 2]
+        num_res_blocks: 2
+        channel_mult: [1, 2, 4]
+        num_head_channels: 64
+        use_spatial_transformer: True
+        use_linear_in_transformer: True
+        transformer_depth: [1, 2, 10]  # note: the first is unused (due to attn_res starting at 2) 32, 16, 8 --> 64, 32, 16
+        context_dim: 2048
+        spatial_transformer_attn_type: softmax-xformers
+        legacy: False
+
+    conditioner_config:
+      target: sgm.modules.GeneralConditioner
+      params:
+        emb_models:
+          # crossattn cond
+          - is_trainable: False
+            input_key: txt
+            target: sgm.modules.encoders.modules.FrozenCLIPEmbedder
+            params:
+              layer: hidden
+              layer_idx: 11
+          # crossattn and vector cond
+          - is_trainable: False
+            input_key: txt
+            target: sgm.modules.encoders.modules.FrozenOpenCLIPEmbedder2
+            params:
+              arch: ViT-bigG-14
+              version: laion2b_s39b_b160k
+              freeze: True
+              layer: penultimate
+              always_return_pooled: True
+              legacy: False
+          # vector cond
+          - is_trainable: False
+            input_key: original_size_as_tuple
+            target: sgm.modules.encoders.modules.ConcatTimestepEmbedderND
+            params:
+              outdim: 256  # multiplied by two
+          # vector cond
+          - is_trainable: False
+            input_key: crop_coords_top_left
+            target: sgm.modules.encoders.modules.ConcatTimestepEmbedderND
+            params:
+              outdim: 256  # multiplied by two
+          # vector cond
+          - is_trainable: False
+            input_key: target_size_as_tuple
+            target: sgm.modules.encoders.modules.ConcatTimestepEmbedderND
+            params:
+              outdim: 256  # multiplied by two
+
+    first_stage_config:
+      target: sgm.models.autoencoder.AutoencoderKLInferenceWrapper
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          attn_type: vanilla-xformers
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult: [1, 2, 4, 4]
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity

configs/v1-inference.yaml

@@ -0,0 +1,70 @@
+model:
+  base_learning_rate: 1.0e-04
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: "jpg"
+    cond_stage_key: "txt"
+    image_size: 64
+    channels: 4
+    cond_stage_trainable: false   # Note: different from the one we trained before
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+
+    scheduler_config: # 10000 warmup steps
+      target: ldm.lr_scheduler.LambdaLinearScheduler
+      params:
+        warm_up_steps: [ 10000 ]
+        cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases
+        f_start: [ 1.e-6 ]
+        f_max: [ 1. ]
+        f_min: [ 1. ]
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32 # unused
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_heads: 8
+        use_spatial_transformer: True
+        transformer_depth: 1
+        context_dim: 768
+        use_checkpoint: True
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder

configs/v1-inpainting-inference.yaml

@@ -0,0 +1,70 @@
+model:
+  base_learning_rate: 7.5e-05
+  target: ldm.models.diffusion.ddpm.LatentInpaintDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: "jpg"
+    cond_stage_key: "txt"
+    image_size: 64
+    channels: 4
+    cond_stage_trainable: false   # Note: different from the one we trained before
+    conditioning_key: hybrid   # important
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    finetune_keys: null
+
+    scheduler_config: # 10000 warmup steps
+      target: ldm.lr_scheduler.LambdaLinearScheduler
+      params:
+        warm_up_steps: [ 2500 ] # NOTE for resuming. use 10000 if starting from scratch
+        cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases
+        f_start: [ 1.e-6 ]
+        f_max: [ 1. ]
+        f_min: [ 1. ]
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32 # unused
+        in_channels: 9  # 4 data + 4 downscaled image + 1 mask
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_heads: 8
+        use_spatial_transformer: True
+        transformer_depth: 1
+        context_dim: 768
+        use_checkpoint: True
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder

environment-wsl2.yaml

@@ -0,0 +1,11 @@
+name: automatic
+channels:
+  - pytorch
+  - defaults
+dependencies:
+  - python=3.10
+  - pip=23.0
+  - cudatoolkit=11.8
+  - pytorch=2.0
+  - torchvision=0.15
+  - numpy=1.23

extensions-builtin/LDSR/ldsr_model_arch.py

@@ -0,0 +1,250 @@
+import os
+import gc
+import time
+
+import numpy as np
+import torch
+import torchvision
+from PIL import Image
+from einops import rearrange, repeat
+from omegaconf import OmegaConf
+import safetensors.torch
+
+from ldm.models.diffusion.ddim import DDIMSampler
+from ldm.util import instantiate_from_config, ismap
+from modules import shared, sd_hijack, devices
+
+cached_ldsr_model: torch.nn.Module = None
+
+
+# Create LDSR Class
+class LDSR:
+    def load_model_from_config(self, half_attention):
+        global cached_ldsr_model
+
+        if shared.opts.ldsr_cached and cached_ldsr_model is not None:
+            print("Loading model from cache")
+            model: torch.nn.Module = cached_ldsr_model
+        else:
+            print(f"Loading model from {self.modelPath}")
+            _, extension = os.path.splitext(self.modelPath)
+            if extension.lower() == ".safetensors":
+                pl_sd = safetensors.torch.load_file(self.modelPath, device="cpu")
+            else:
+                pl_sd = torch.load(self.modelPath, map_location="cpu")
+            sd = pl_sd["state_dict"] if "state_dict" in pl_sd else pl_sd
+            config = OmegaConf.load(self.yamlPath)
+            config.model.target = "ldm.models.diffusion.ddpm.LatentDiffusionV1"
+            model: torch.nn.Module = instantiate_from_config(config.model)
+            model.load_state_dict(sd, strict=False)
+            model = model.to(shared.device)
+            if half_attention:
+                model = model.half()
+            if shared.cmd_opts.opt_channelslast:
+                model = model.to(memory_format=torch.channels_last)
+
+            sd_hijack.model_hijack.hijack(model) # apply optimization
+            model.eval()
+
+            if shared.opts.ldsr_cached:
+                cached_ldsr_model = model
+
+        return {"model": model}
+
+    def __init__(self, model_path, yaml_path):
+        self.modelPath = model_path
+        self.yamlPath = yaml_path
+
+    @staticmethod
+    def run(model, selected_path, custom_steps, eta):
+        example = get_cond(selected_path)
+
+        n_runs = 1
+        guider = None
+        ckwargs = None
+        ddim_use_x0_pred = False
+        temperature = 1.
+        eta = eta
+        custom_shape = None
+
+        height, width = example["image"].shape[1:3]
+        split_input = height >= 128 and width >= 128
+
+        if split_input:
+            ks = 128
+            stride = 64
+            vqf = 4  #
+            model.split_input_params = {"ks": (ks, ks), "stride": (stride, stride),
+                                        "vqf": vqf,
+                                        "patch_distributed_vq": True,
+                                        "tie_braker": False,
+                                        "clip_max_weight": 0.5,
+                                        "clip_min_weight": 0.01,
+                                        "clip_max_tie_weight": 0.5,
+                                        "clip_min_tie_weight": 0.01}
+        else:
+            if hasattr(model, "split_input_params"):
+                delattr(model, "split_input_params")
+
+        x_t = None
+        logs = None
+        for _ in range(n_runs):
+            if custom_shape is not None:
+                x_t = torch.randn(1, custom_shape[1], custom_shape[2], custom_shape[3]).to(model.device)
+                x_t = repeat(x_t, '1 c h w -> b c h w', b=custom_shape[0])
+
+            logs = make_convolutional_sample(example, model,
+                                             custom_steps=custom_steps,
+                                             eta=eta, quantize_x0=False,
+                                             custom_shape=custom_shape,
+                                             temperature=temperature, noise_dropout=0.,
+                                             corrector=guider, corrector_kwargs=ckwargs, x_T=x_t,
+                                             ddim_use_x0_pred=ddim_use_x0_pred
+                                             )
+        return logs
+
+    def super_resolution(self, image, steps=100, target_scale=2, half_attention=False):
+        model = self.load_model_from_config(half_attention)
+
+        # Run settings
+        diffusion_steps = int(steps)
+        eta = 1.0
+
+
+        gc.collect()
+        devices.torch_gc()
+
+        im_og = image
+        width_og, height_og = im_og.size
+        # If we can adjust the max upscale size, then the 4 below should be our variable
+        down_sample_rate = target_scale / 4
+        wd = width_og * down_sample_rate
+        hd = height_og * down_sample_rate
+        width_downsampled_pre = int(np.ceil(wd))
+        height_downsampled_pre = int(np.ceil(hd))
+
+        if down_sample_rate != 1:
+            print(
+                f'Downsampling from [{width_og}, {height_og}] to [{width_downsampled_pre}, {height_downsampled_pre}]')
+            im_og = im_og.resize((width_downsampled_pre, height_downsampled_pre), Image.LANCZOS)
+        else:
+            print(f"Down sample rate is 1 from {target_scale} / 4 (Not downsampling)")
+
+        # pad width and height to multiples of 64, pads with the edge values of image to avoid artifacts
+        pad_w, pad_h = np.max(((2, 2), np.ceil(np.array(im_og.size) / 64).astype(int)), axis=0) * 64 - im_og.size
+        im_padded = Image.fromarray(np.pad(np.array(im_og), ((0, pad_h), (0, pad_w), (0, 0)), mode='edge'))
+
+        logs = self.run(model["model"], im_padded, diffusion_steps, eta)
+
+        sample = logs["sample"]
+        sample = sample.detach().cpu()
+        sample = torch.clamp(sample, -1., 1.)
+        sample = (sample + 1.) / 2. * 255
+        sample = sample.numpy().astype(np.uint8)
+        sample = np.transpose(sample, (0, 2, 3, 1))
+        a = Image.fromarray(sample[0])
+
+        # remove padding
+        a = a.crop((0, 0) + tuple(np.array(im_og.size) * 4))
+
+        del model
+        gc.collect()
+        devices.torch_gc()
+
+        return a
+
+
+def get_cond(selected_path):
+    example = {}
+    up_f = 4
+    c = selected_path.convert('RGB')
+    c = torch.unsqueeze(torchvision.transforms.ToTensor()(c), 0)
+    c_up = torchvision.transforms.functional.resize(c, size=[up_f * c.shape[2], up_f * c.shape[3]],
+                                                    antialias=True)
+    c_up = rearrange(c_up, '1 c h w -> 1 h w c')
+    c = rearrange(c, '1 c h w -> 1 h w c')
+    c = 2. * c - 1.
+
+    c = c.to(shared.device)
+    example["LR_image"] = c
+    example["image"] = c_up
+
+    return example
+
+
+@torch.no_grad()
+def convsample_ddim(model, cond, steps, shape, eta=1.0, callback=None, normals_sequence=None,
+                    mask=None, x0=None, quantize_x0=False, temperature=1., score_corrector=None,
+                    corrector_kwargs=None, x_t=None
+                    ):
+    ddim = DDIMSampler(model)
+    bs = shape[0]
+    shape = shape[1:]
+    print(f"Sampling with eta = {eta}; steps: {steps}")
+    samples, intermediates = ddim.sample(steps, batch_size=bs, shape=shape, conditioning=cond, callback=callback,
+                                         normals_sequence=normals_sequence, quantize_x0=quantize_x0, eta=eta,
+                                         mask=mask, x0=x0, temperature=temperature, verbose=False,
+                                         score_corrector=score_corrector,
+                                         corrector_kwargs=corrector_kwargs, x_t=x_t)
+
+    return samples, intermediates
+
+
+@torch.no_grad()
+def make_convolutional_sample(batch, model, custom_steps=None, eta=1.0, quantize_x0=False, custom_shape=None, temperature=1., noise_dropout=0., corrector=None,
+                              corrector_kwargs=None, x_T=None, ddim_use_x0_pred=False):
+    log = {}
+
+    z, c, x, xrec, xc = model.get_input(batch, model.first_stage_key,
+                                        return_first_stage_outputs=True,
+                                        force_c_encode=not (hasattr(model, 'split_input_params')
+                                                            and model.cond_stage_key == 'coordinates_bbox'),
+                                        return_original_cond=True)
+
+    if custom_shape is not None:
+        z = torch.randn(custom_shape)
+        print(f"Generating {custom_shape[0]} samples of shape {custom_shape[1:]}")
+
+    z0 = None
+
+    log["input"] = x
+    log["reconstruction"] = xrec
+
+    if ismap(xc):
+        log["original_conditioning"] = model.to_rgb(xc)
+        if hasattr(model, 'cond_stage_key'):
+            log[model.cond_stage_key] = model.to_rgb(xc)
+
+    else:
+        log["original_conditioning"] = xc if xc is not None else torch.zeros_like(x)
+        if model.cond_stage_model:
+            log[model.cond_stage_key] = xc if xc is not None else torch.zeros_like(x)
+            if model.cond_stage_key == 'class_label':
+                log[model.cond_stage_key] = xc[model.cond_stage_key]
+
+    with model.ema_scope("Plotting"):
+        t0 = time.time()
+
+        sample, intermediates = convsample_ddim(model, c, steps=custom_steps, shape=z.shape,
+                                                eta=eta,
+                                                quantize_x0=quantize_x0, mask=None, x0=z0,
+                                                temperature=temperature, score_corrector=corrector, corrector_kwargs=corrector_kwargs,
+                                                x_t=x_T)
+        t1 = time.time()
+
+        if ddim_use_x0_pred:
+            sample = intermediates['pred_x0'][-1]
+
+    x_sample = model.decode_first_stage(sample)
+
+    try:
+        x_sample_noquant = model.decode_first_stage(sample, force_not_quantize=True)
+        log["sample_noquant"] = x_sample_noquant
+        log["sample_diff"] = torch.abs(x_sample_noquant - x_sample)
+    except Exception:
+        pass
+
+    log["sample"] = x_sample
+    log["time"] = t1 - t0
+
+    return log

extensions-builtin/LDSR/preload.py

@@ -0,0 +1,6 @@
+import os
+from modules import paths
+
+
+def preload(parser):
+    parser.add_argument("--ldsr-models-path", type=str, help="Path to directory with LDSR model file(s).", default=os.path.join(paths.models_path, 'LDSR'))

extensions-builtin/LDSR/scripts/ldsr_model.py

@@ -0,0 +1,70 @@
+import os
+
+from modules.modelloader import load_file_from_url
+from modules.upscaler import Upscaler, UpscalerData
+from modules_forge.forge_util import prepare_free_memory
+from ldsr_model_arch import LDSR
+from modules import shared, script_callbacks, errors
+import sd_hijack_autoencoder  # noqa: F401
+import sd_hijack_ddpm_v1  # noqa: F401
+
+
+class UpscalerLDSR(Upscaler):
+    def __init__(self, user_path):
+        self.name = "LDSR"
+        self.user_path = user_path
+        self.model_url = "https://heibox.uni-heidelberg.de/f/578df07c8fc04ffbadf3/?dl=1"
+        self.yaml_url = "https://heibox.uni-heidelberg.de/f/31a76b13ea27482981b4/?dl=1"
+        super().__init__()
+        scaler_data = UpscalerData("LDSR", None, self)
+        self.scalers = [scaler_data]
+
+    def load_model(self, path: str):
+        # Remove incorrect project.yaml file if too big
+        yaml_path = os.path.join(self.model_path, "project.yaml")
+        old_model_path = os.path.join(self.model_path, "model.pth")
+        new_model_path = os.path.join(self.model_path, "model.ckpt")
+
+        local_model_paths = self.find_models(ext_filter=[".ckpt", ".safetensors"])
+        local_ckpt_path = next(iter([local_model for local_model in local_model_paths if local_model.endswith("model.ckpt")]), None)
+        local_safetensors_path = next(iter([local_model for local_model in local_model_paths if local_model.endswith("model.safetensors")]), None)
+        local_yaml_path = next(iter([local_model for local_model in local_model_paths if local_model.endswith("project.yaml")]), None)
+
+        if os.path.exists(yaml_path):
+            statinfo = os.stat(yaml_path)
+            if statinfo.st_size >= 10485760:
+                print("Removing invalid LDSR YAML file.")
+                os.remove(yaml_path)
+
+        if os.path.exists(old_model_path):
+            print("Renaming model from model.pth to model.ckpt")
+            os.rename(old_model_path, new_model_path)
+
+        if local_safetensors_path is not None and os.path.exists(local_safetensors_path):
+            model = local_safetensors_path
+        else:
+            model = local_ckpt_path or load_file_from_url(self.model_url, model_dir=self.model_download_path, file_name="model.ckpt")
+
+        yaml = local_yaml_path or load_file_from_url(self.yaml_url, model_dir=self.model_download_path, file_name="project.yaml")
+
+        return LDSR(model, yaml)
+
+    def do_upscale(self, img, path):
+        prepare_free_memory(aggressive=True)
+        try:
+            ldsr = self.load_model(path)
+        except Exception:
+            errors.report(f"Failed loading LDSR model {path}", exc_info=True)
+            return img
+        ddim_steps = shared.opts.ldsr_steps
+        return ldsr.super_resolution(img, ddim_steps, self.scale)
+
+
+def on_ui_settings():
+    import gradio as gr
+
+    shared.opts.add_option("ldsr_steps", shared.OptionInfo(100, "LDSR processing steps. Lower = faster", gr.Slider, {"minimum": 1, "maximum": 200, "step": 1}, section=('upscaling', "Upscaling")))
+    shared.opts.add_option("ldsr_cached", shared.OptionInfo(False, "Cache LDSR model in memory", gr.Checkbox, {"interactive": True}, section=('upscaling', "Upscaling")))
+
+
+script_callbacks.on_ui_settings(on_ui_settings)

extensions-builtin/LDSR/sd_hijack_autoencoder.py

@@ -0,0 +1,293 @@
+# The content of this file comes from the ldm/models/autoencoder.py file of the compvis/stable-diffusion repo
+# The VQModel & VQModelInterface were subsequently removed from ldm/models/autoencoder.py when we moved to the stability-ai/stablediffusion repo
+# As the LDSR upscaler relies on VQModel & VQModelInterface, the hijack aims to put them back into the ldm.models.autoencoder
+import numpy as np
+import torch
+import pytorch_lightning as pl
+import torch.nn.functional as F
+from contextlib import contextmanager
+
+from torch.optim.lr_scheduler import LambdaLR
+
+from ldm.modules.ema import LitEma
+from vqvae_quantize import VectorQuantizer2 as VectorQuantizer
+from ldm.modules.diffusionmodules.model import Encoder, Decoder
+from ldm.util import instantiate_from_config
+
+import ldm.models.autoencoder
+from packaging import version
+
+class VQModel(pl.LightningModule):
+    def __init__(self,
+                 ddconfig,
+                 lossconfig,
+                 n_embed,
+                 embed_dim,
+                 ckpt_path=None,
+                 ignore_keys=None,
+                 image_key="image",
+                 colorize_nlabels=None,
+                 monitor=None,
+                 batch_resize_range=None,
+                 scheduler_config=None,
+                 lr_g_factor=1.0,
+                 remap=None,
+                 sane_index_shape=False, # tell vector quantizer to return indices as bhw
+                 use_ema=False
+                 ):
+        super().__init__()
+        self.embed_dim = embed_dim
+        self.n_embed = n_embed
+        self.image_key = image_key
+        self.encoder = Encoder(**ddconfig)
+        self.decoder = Decoder(**ddconfig)
+        self.loss = instantiate_from_config(lossconfig)
+        self.quantize = VectorQuantizer(n_embed, embed_dim, beta=0.25,
+                                        remap=remap,
+                                        sane_index_shape=sane_index_shape)
+        self.quant_conv = torch.nn.Conv2d(ddconfig["z_channels"], embed_dim, 1)
+        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
+        if colorize_nlabels is not None:
+            assert type(colorize_nlabels)==int
+            self.register_buffer("colorize", torch.randn(3, colorize_nlabels, 1, 1))
+        if monitor is not None:
+            self.monitor = monitor
+        self.batch_resize_range = batch_resize_range
+        if self.batch_resize_range is not None:
+            print(f"{self.__class__.__name__}: Using per-batch resizing in range {batch_resize_range}.")
+
+        self.use_ema = use_ema
+        if self.use_ema:
+            self.model_ema = LitEma(self)
+            print(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.")
+
+        if ckpt_path is not None:
+            self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys or [])
+        self.scheduler_config = scheduler_config
+        self.lr_g_factor = lr_g_factor
+
+    @contextmanager
+    def ema_scope(self, context=None):
+        if self.use_ema:
+            self.model_ema.store(self.parameters())
+            self.model_ema.copy_to(self)
+            if context is not None:
+                print(f"{context}: Switched to EMA weights")
+        try:
+            yield None
+        finally:
+            if self.use_ema:
+                self.model_ema.restore(self.parameters())
+                if context is not None:
+                    print(f"{context}: Restored training weights")
+
+    def init_from_ckpt(self, path, ignore_keys=None):
+        sd = torch.load(path, map_location="cpu")["state_dict"]
+        keys = list(sd.keys())
+        for k in keys:
+            for ik in ignore_keys or []:
+                if k.startswith(ik):
+                    print("Deleting key {} from state_dict.".format(k))
+                    del sd[k]
+        missing, unexpected = self.load_state_dict(sd, strict=False)
+        print(f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys")
+        if missing:
+            print(f"Missing Keys: {missing}")
+        if unexpected:
+            print(f"Unexpected Keys: {unexpected}")
+
+    def on_train_batch_end(self, *args, **kwargs):
+        if self.use_ema:
+            self.model_ema(self)
+
+    def encode(self, x):
+        h = self.encoder(x)
+        h = self.quant_conv(h)
+        quant, emb_loss, info = self.quantize(h)
+        return quant, emb_loss, info
+
+    def encode_to_prequant(self, x):
+        h = self.encoder(x)
+        h = self.quant_conv(h)
+        return h
+
+    def decode(self, quant):
+        quant = self.post_quant_conv(quant)
+        dec = self.decoder(quant)
+        return dec
+
+    def decode_code(self, code_b):
+        quant_b = self.quantize.embed_code(code_b)
+        dec = self.decode(quant_b)
+        return dec
+
+    def forward(self, input, return_pred_indices=False):
+        quant, diff, (_,_,ind) = self.encode(input)
+        dec = self.decode(quant)
+        if return_pred_indices:
+            return dec, diff, ind
+        return dec, diff
+
+    def get_input(self, batch, k):
+        x = batch[k]
+        if len(x.shape) == 3:
+            x = x[..., None]
+        x = x.permute(0, 3, 1, 2).to(memory_format=torch.contiguous_format).float()
+        if self.batch_resize_range is not None:
+            lower_size = self.batch_resize_range[0]
+            upper_size = self.batch_resize_range[1]
+            if self.global_step <= 4:
+                # do the first few batches with max size to avoid later oom
+                new_resize = upper_size
+            else:
+                new_resize = np.random.choice(np.arange(lower_size, upper_size+16, 16))
+            if new_resize != x.shape[2]:
+                x = F.interpolate(x, size=new_resize, mode="bicubic")
+            x = x.detach()
+        return x
+
+    def training_step(self, batch, batch_idx, optimizer_idx):
+        # https://github.com/pytorch/pytorch/issues/37142
+        # try not to fool the heuristics
+        x = self.get_input(batch, self.image_key)
+        xrec, qloss, ind = self(x, return_pred_indices=True)
+
+        if optimizer_idx == 0:
+            # autoencode
+            aeloss, log_dict_ae = self.loss(qloss, x, xrec, optimizer_idx, self.global_step,
+                                            last_layer=self.get_last_layer(), split="train",
+                                            predicted_indices=ind)
+
+            self.log_dict(log_dict_ae, prog_bar=False, logger=True, on_step=True, on_epoch=True)
+            return aeloss
+
+        if optimizer_idx == 1:
+            # discriminator
+            discloss, log_dict_disc = self.loss(qloss, x, xrec, optimizer_idx, self.global_step,
+                                            last_layer=self.get_last_layer(), split="train")
+            self.log_dict(log_dict_disc, prog_bar=False, logger=True, on_step=True, on_epoch=True)
+            return discloss
+
+    def validation_step(self, batch, batch_idx):
+        log_dict = self._validation_step(batch, batch_idx)
+        with self.ema_scope():
+            self._validation_step(batch, batch_idx, suffix="_ema")
+        return log_dict
+
+    def _validation_step(self, batch, batch_idx, suffix=""):
+        x = self.get_input(batch, self.image_key)
+        xrec, qloss, ind = self(x, return_pred_indices=True)
+        aeloss, log_dict_ae = self.loss(qloss, x, xrec, 0,
+                                        self.global_step,
+                                        last_layer=self.get_last_layer(),
+                                        split="val"+suffix,
+                                        predicted_indices=ind
+                                        )
+
+        discloss, log_dict_disc = self.loss(qloss, x, xrec, 1,
+                                            self.global_step,
+                                            last_layer=self.get_last_layer(),
+                                            split="val"+suffix,
+                                            predicted_indices=ind
+                                            )
+        rec_loss = log_dict_ae[f"val{suffix}/rec_loss"]
+        self.log(f"val{suffix}/rec_loss", rec_loss,
+                   prog_bar=True, logger=True, on_step=False, on_epoch=True, sync_dist=True)
+        self.log(f"val{suffix}/aeloss", aeloss,
+                   prog_bar=True, logger=True, on_step=False, on_epoch=True, sync_dist=True)
+        if version.parse(pl.__version__) >= version.parse('1.4.0'):
+            del log_dict_ae[f"val{suffix}/rec_loss"]
+        self.log_dict(log_dict_ae)
+        self.log_dict(log_dict_disc)
+        return self.log_dict
+
+    def configure_optimizers(self):
+        lr_d = self.learning_rate
+        lr_g = self.lr_g_factor*self.learning_rate
+        print("lr_d", lr_d)
+        print("lr_g", lr_g)
+        opt_ae = torch.optim.Adam(list(self.encoder.parameters())+
+                                  list(self.decoder.parameters())+
+                                  list(self.quantize.parameters())+
+                                  list(self.quant_conv.parameters())+
+                                  list(self.post_quant_conv.parameters()),
+                                  lr=lr_g, betas=(0.5, 0.9))
+        opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(),
+                                    lr=lr_d, betas=(0.5, 0.9))
+
+        if self.scheduler_config is not None:
+            scheduler = instantiate_from_config(self.scheduler_config)
+
+            print("Setting up LambdaLR scheduler...")
+            scheduler = [
+                {
+                    'scheduler': LambdaLR(opt_ae, lr_lambda=scheduler.schedule),
+                    'interval': 'step',
+                    'frequency': 1
+                },
+                {
+                    'scheduler': LambdaLR(opt_disc, lr_lambda=scheduler.schedule),
+                    'interval': 'step',
+                    'frequency': 1
+                },
+            ]
+            return [opt_ae, opt_disc], scheduler
+        return [opt_ae, opt_disc], []
+
+    def get_last_layer(self):
+        return self.decoder.conv_out.weight
+
+    def log_images(self, batch, only_inputs=False, plot_ema=False, **kwargs):
+        log = {}
+        x = self.get_input(batch, self.image_key)
+        x = x.to(self.device)
+        if only_inputs:
+            log["inputs"] = x
+            return log
+        xrec, _ = self(x)
+        if x.shape[1] > 3:
+            # colorize with random projection
+            assert xrec.shape[1] > 3
+            x = self.to_rgb(x)
+            xrec = self.to_rgb(xrec)
+        log["inputs"] = x
+        log["reconstructions"] = xrec
+        if plot_ema:
+            with self.ema_scope():
+                xrec_ema, _ = self(x)
+                if x.shape[1] > 3:
+                    xrec_ema = self.to_rgb(xrec_ema)
+                log["reconstructions_ema"] = xrec_ema
+        return log
+
+    def to_rgb(self, x):
+        assert self.image_key == "segmentation"
+        if not hasattr(self, "colorize"):
+            self.register_buffer("colorize", torch.randn(3, x.shape[1], 1, 1).to(x))
+        x = F.conv2d(x, weight=self.colorize)
+        x = 2.*(x-x.min())/(x.max()-x.min()) - 1.
+        return x
+
+
+class VQModelInterface(VQModel):
+    def __init__(self, embed_dim, *args, **kwargs):
+        super().__init__(*args, embed_dim=embed_dim, **kwargs)
+        self.embed_dim = embed_dim
+
+    def encode(self, x):
+        h = self.encoder(x)
+        h = self.quant_conv(h)
+        return h
+
+    def decode(self, h, force_not_quantize=False):
+        # also go through quantization layer
+        if not force_not_quantize:
+            quant, emb_loss, info = self.quantize(h)
+        else:
+            quant = h
+        quant = self.post_quant_conv(quant)
+        dec = self.decoder(quant)
+        return dec
+
+ldm.models.autoencoder.VQModel = VQModel
+ldm.models.autoencoder.VQModelInterface = VQModelInterface

extensions-builtin/LDSR/sd_hijack_ddpm_v1.py

@@ -0,0 +1,1443 @@
+# This script is copied from the compvis/stable-diffusion repo (aka the SD V1 repo)
+# Original filename: ldm/models/diffusion/ddpm.py
+# The purpose to reinstate the old DDPM logic which works with VQ, whereas the V2 one doesn't
+# Some models such as LDSR require VQ to work correctly
+# The classes are suffixed with "V1" and added back to the "ldm.models.diffusion.ddpm" module
+
+import torch
+import torch.nn as nn
+import numpy as np
+import pytorch_lightning as pl
+from torch.optim.lr_scheduler import LambdaLR
+from einops import rearrange, repeat
+from contextlib import contextmanager
+from functools import partial
+from tqdm import tqdm
+from torchvision.utils import make_grid
+from pytorch_lightning.utilities.distributed import rank_zero_only
+
+from ldm.util import log_txt_as_img, exists, default, ismap, isimage, mean_flat, count_params, instantiate_from_config
+from ldm.modules.ema import LitEma
+from ldm.modules.distributions.distributions import normal_kl, DiagonalGaussianDistribution
+from ldm.models.autoencoder import VQModelInterface, IdentityFirstStage, AutoencoderKL
+from ldm.modules.diffusionmodules.util import make_beta_schedule, extract_into_tensor, noise_like
+from ldm.models.diffusion.ddim import DDIMSampler
+
+import ldm.models.diffusion.ddpm
+
+__conditioning_keys__ = {'concat': 'c_concat',
+                         'crossattn': 'c_crossattn',
+                         'adm': 'y'}
+
+
+def disabled_train(self, mode=True):
+    """Overwrite model.train with this function to make sure train/eval mode
+    does not change anymore."""
+    return self
+
+
+def uniform_on_device(r1, r2, shape, device):
+    return (r1 - r2) * torch.rand(*shape, device=device) + r2
+
+
+class DDPMV1(pl.LightningModule):
+    # classic DDPM with Gaussian diffusion, in image space
+    def __init__(self,
+                 unet_config,
+                 timesteps=1000,
+                 beta_schedule="linear",
+                 loss_type="l2",
+                 ckpt_path=None,
+                 ignore_keys=None,
+                 load_only_unet=False,
+                 monitor="val/loss",
+                 use_ema=True,
+                 first_stage_key="image",
+                 image_size=256,
+                 channels=3,
+                 log_every_t=100,
+                 clip_denoised=True,
+                 linear_start=1e-4,
+                 linear_end=2e-2,
+                 cosine_s=8e-3,
+                 given_betas=None,
+                 original_elbo_weight=0.,
+                 v_posterior=0.,  # weight for choosing posterior variance as sigma = (1-v) * beta_tilde + v * beta
+                 l_simple_weight=1.,
+                 conditioning_key=None,
+                 parameterization="eps",  # all assuming fixed variance schedules
+                 scheduler_config=None,
+                 use_positional_encodings=False,
+                 learn_logvar=False,
+                 logvar_init=0.,
+                 ):
+        super().__init__()
+        assert parameterization in ["eps", "x0"], 'currently only supporting "eps" and "x0"'
+        self.parameterization = parameterization
+        print(f"{self.__class__.__name__}: Running in {self.parameterization}-prediction mode")
+        self.cond_stage_model = None
+        self.clip_denoised = clip_denoised
+        self.log_every_t = log_every_t
+        self.first_stage_key = first_stage_key
+        self.image_size = image_size  # try conv?
+        self.channels = channels
+        self.use_positional_encodings = use_positional_encodings
+        self.model = DiffusionWrapperV1(unet_config, conditioning_key)
+        count_params(self.model, verbose=True)
+        self.use_ema = use_ema
+        if self.use_ema:
+            self.model_ema = LitEma(self.model)
+            print(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.")
+
+        self.use_scheduler = scheduler_config is not None
+        if self.use_scheduler:
+            self.scheduler_config = scheduler_config
+
+        self.v_posterior = v_posterior
+        self.original_elbo_weight = original_elbo_weight
+        self.l_simple_weight = l_simple_weight
+
+        if monitor is not None:
+            self.monitor = monitor
+        if ckpt_path is not None:
+            self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys or [], only_model=load_only_unet)
+
+        self.register_schedule(given_betas=given_betas, beta_schedule=beta_schedule, timesteps=timesteps,
+                               linear_start=linear_start, linear_end=linear_end, cosine_s=cosine_s)
+
+        self.loss_type = loss_type
+
+        self.learn_logvar = learn_logvar
+        self.logvar = torch.full(fill_value=logvar_init, size=(self.num_timesteps,))
+        if self.learn_logvar:
+            self.logvar = nn.Parameter(self.logvar, requires_grad=True)
+
+
+    def register_schedule(self, given_betas=None, beta_schedule="linear", timesteps=1000,
+                          linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
+        if exists(given_betas):
+            betas = given_betas
+        else:
+            betas = make_beta_schedule(beta_schedule, timesteps, linear_start=linear_start, linear_end=linear_end,
+                                       cosine_s=cosine_s)
+        alphas = 1. - betas
+        alphas_cumprod = np.cumprod(alphas, axis=0)
+        alphas_cumprod_prev = np.append(1., alphas_cumprod[:-1])
+
+        timesteps, = betas.shape
+        self.num_timesteps = int(timesteps)
+        self.linear_start = linear_start
+        self.linear_end = linear_end
+        assert alphas_cumprod.shape[0] == self.num_timesteps, 'alphas have to be defined for each timestep'
+
+        to_torch = partial(torch.tensor, dtype=torch.float32)
+
+        self.register_buffer('betas', to_torch(betas))
+        self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
+        self.register_buffer('alphas_cumprod_prev', to_torch(alphas_cumprod_prev))
+
+        # calculations for diffusion q(x_t | x_{t-1}) and others
+        self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod)))
+        self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod)))
+        self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod)))
+        self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod)))
+        self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod - 1)))
+
+        # calculations for posterior q(x_{t-1} | x_t, x_0)
+        posterior_variance = (1 - self.v_posterior) * betas * (1. - alphas_cumprod_prev) / (
+                    1. - alphas_cumprod) + self.v_posterior * betas
+        # above: equal to 1. / (1. / (1. - alpha_cumprod_tm1) + alpha_t / beta_t)
+        self.register_buffer('posterior_variance', to_torch(posterior_variance))
+        # below: log calculation clipped because the posterior variance is 0 at the beginning of the diffusion chain
+        self.register_buffer('posterior_log_variance_clipped', to_torch(np.log(np.maximum(posterior_variance, 1e-20))))
+        self.register_buffer('posterior_mean_coef1', to_torch(
+            betas * np.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod)))
+        self.register_buffer('posterior_mean_coef2', to_torch(
+            (1. - alphas_cumprod_prev) * np.sqrt(alphas) / (1. - alphas_cumprod)))
+
+        if self.parameterization == "eps":
+            lvlb_weights = self.betas ** 2 / (
+                        2 * self.posterior_variance * to_torch(alphas) * (1 - self.alphas_cumprod))
+        elif self.parameterization == "x0":
+            lvlb_weights = 0.5 * np.sqrt(torch.Tensor(alphas_cumprod)) / (2. * 1 - torch.Tensor(alphas_cumprod))
+        else:
+            raise NotImplementedError("mu not supported")
+        # TODO how to choose this term
+        lvlb_weights[0] = lvlb_weights[1]
+        self.register_buffer('lvlb_weights', lvlb_weights, persistent=False)
+        assert not torch.isnan(self.lvlb_weights).all()
+
+    @contextmanager
+    def ema_scope(self, context=None):
+        if self.use_ema:
+            self.model_ema.store(self.model.parameters())
+            self.model_ema.copy_to(self.model)
+            if context is not None:
+                print(f"{context}: Switched to EMA weights")
+        try:
+            yield None
+        finally:
+            if self.use_ema:
+                self.model_ema.restore(self.model.parameters())
+                if context is not None:
+                    print(f"{context}: Restored training weights")
+
+    def init_from_ckpt(self, path, ignore_keys=None, only_model=False):
+        sd = torch.load(path, map_location="cpu")
+        if "state_dict" in list(sd.keys()):
+            sd = sd["state_dict"]
+        keys = list(sd.keys())
+        for k in keys:
+            for ik in ignore_keys or []:
+                if k.startswith(ik):
+                    print("Deleting key {} from state_dict.".format(k))
+                    del sd[k]
+        missing, unexpected = self.load_state_dict(sd, strict=False) if not only_model else self.model.load_state_dict(
+            sd, strict=False)
+        print(f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys")
+        if missing:
+            print(f"Missing Keys: {missing}")
+        if unexpected:
+            print(f"Unexpected Keys: {unexpected}")
+
+    def q_mean_variance(self, x_start, t):
+        """
+        Get the distribution q(x_t | x_0).
+        :param x_start: the [N x C x ...] tensor of noiseless inputs.
+        :param t: the number of diffusion steps (minus 1). Here, 0 means one step.
+        :return: A tuple (mean, variance, log_variance), all of x_start's shape.
+        """
+        mean = (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start)
+        variance = extract_into_tensor(1.0 - self.alphas_cumprod, t, x_start.shape)
+        log_variance = extract_into_tensor(self.log_one_minus_alphas_cumprod, t, x_start.shape)
+        return mean, variance, log_variance
+
+    def predict_start_from_noise(self, x_t, t, noise):
+        return (
+                extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t -
+                extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise
+        )
+
+    def q_posterior(self, x_start, x_t, t):
+        posterior_mean = (
+                extract_into_tensor(self.posterior_mean_coef1, t, x_t.shape) * x_start +
+                extract_into_tensor(self.posterior_mean_coef2, t, x_t.shape) * x_t
+        )
+        posterior_variance = extract_into_tensor(self.posterior_variance, t, x_t.shape)
+        posterior_log_variance_clipped = extract_into_tensor(self.posterior_log_variance_clipped, t, x_t.shape)
+        return posterior_mean, posterior_variance, posterior_log_variance_clipped
+
+    def p_mean_variance(self, x, t, clip_denoised: bool):
+        model_out = self.model(x, t)
+        if self.parameterization == "eps":
+            x_recon = self.predict_start_from_noise(x, t=t, noise=model_out)
+        elif self.parameterization == "x0":
+            x_recon = model_out
+        if clip_denoised:
+            x_recon.clamp_(-1., 1.)
+
+        model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
+        return model_mean, posterior_variance, posterior_log_variance
+
+    @torch.no_grad()
+    def p_sample(self, x, t, clip_denoised=True, repeat_noise=False):
+        b, *_, device = *x.shape, x.device
+        model_mean, _, model_log_variance = self.p_mean_variance(x=x, t=t, clip_denoised=clip_denoised)
+        noise = noise_like(x.shape, device, repeat_noise)
+        # no noise when t == 0
+        nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
+        return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
+
+    @torch.no_grad()
+    def p_sample_loop(self, shape, return_intermediates=False):
+        device = self.betas.device
+        b = shape[0]
+        img = torch.randn(shape, device=device)
+        intermediates = [img]
+        for i in tqdm(reversed(range(0, self.num_timesteps)), desc='Sampling t', total=self.num_timesteps):
+            img = self.p_sample(img, torch.full((b,), i, device=device, dtype=torch.long),
+                                clip_denoised=self.clip_denoised)
+            if i % self.log_every_t == 0 or i == self.num_timesteps - 1:
+                intermediates.append(img)
+        if return_intermediates:
+            return img, intermediates
+        return img
+
+    @torch.no_grad()
+    def sample(self, batch_size=16, return_intermediates=False):
+        image_size = self.image_size
+        channels = self.channels
+        return self.p_sample_loop((batch_size, channels, image_size, image_size),
+                                  return_intermediates=return_intermediates)
+
+    def q_sample(self, x_start, t, noise=None):
+        noise = default(noise, lambda: torch.randn_like(x_start))
+        return (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start +
+                extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise)
+
+    def get_loss(self, pred, target, mean=True):
+        if self.loss_type == 'l1':
+            loss = (target - pred).abs()
+            if mean:
+                loss = loss.mean()
+        elif self.loss_type == 'l2':
+            if mean:
+                loss = torch.nn.functional.mse_loss(target, pred)
+            else:
+                loss = torch.nn.functional.mse_loss(target, pred, reduction='none')
+        else:
+            raise NotImplementedError("unknown loss type '{loss_type}'")
+
+        return loss
+
+    def p_losses(self, x_start, t, noise=None):
+        noise = default(noise, lambda: torch.randn_like(x_start))
+        x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
+        model_out = self.model(x_noisy, t)
+
+        loss_dict = {}
+        if self.parameterization == "eps":
+            target = noise
+        elif self.parameterization == "x0":
+            target = x_start
+        else:
+            raise NotImplementedError(f"Paramterization {self.parameterization} not yet supported")
+
+        loss = self.get_loss(model_out, target, mean=False).mean(dim=[1, 2, 3])
+
+        log_prefix = 'train' if self.training else 'val'
+
+        loss_dict.update({f'{log_prefix}/loss_simple': loss.mean()})
+        loss_simple = loss.mean() * self.l_simple_weight
+
+        loss_vlb = (self.lvlb_weights[t] * loss).mean()
+        loss_dict.update({f'{log_prefix}/loss_vlb': loss_vlb})
+
+        loss = loss_simple + self.original_elbo_weight * loss_vlb
+
+        loss_dict.update({f'{log_prefix}/loss': loss})
+
+        return loss, loss_dict
+
+    def forward(self, x, *args, **kwargs):
+        # b, c, h, w, device, img_size, = *x.shape, x.device, self.image_size
+        # assert h == img_size and w == img_size, f'height and width of image must be {img_size}'
+        t = torch.randint(0, self.num_timesteps, (x.shape[0],), device=self.device).long()
+        return self.p_losses(x, t, *args, **kwargs)
+
+    def get_input(self, batch, k):
+        x = batch[k]
+        if len(x.shape) == 3:
+            x = x[..., None]
+        x = rearrange(x, 'b h w c -> b c h w')
+        x = x.to(memory_format=torch.contiguous_format).float()
+        return x
+
+    def shared_step(self, batch):
+        x = self.get_input(batch, self.first_stage_key)
+        loss, loss_dict = self(x)
+        return loss, loss_dict
+
+    def training_step(self, batch, batch_idx):
+        loss, loss_dict = self.shared_step(batch)
+
+        self.log_dict(loss_dict, prog_bar=True,
+                      logger=True, on_step=True, on_epoch=True)
+
+        self.log("global_step", self.global_step,
+                 prog_bar=True, logger=True, on_step=True, on_epoch=False)
+
+        if self.use_scheduler:
+            lr = self.optimizers().param_groups[0]['lr']
+            self.log('lr_abs', lr, prog_bar=True, logger=True, on_step=True, on_epoch=False)
+
+        return loss
+
+    @torch.no_grad()
+    def validation_step(self, batch, batch_idx):
+        _, loss_dict_no_ema = self.shared_step(batch)
+        with self.ema_scope():
+            _, loss_dict_ema = self.shared_step(batch)
+            loss_dict_ema = {key + '_ema': loss_dict_ema[key] for key in loss_dict_ema}
+        self.log_dict(loss_dict_no_ema, prog_bar=False, logger=True, on_step=False, on_epoch=True)
+        self.log_dict(loss_dict_ema, prog_bar=False, logger=True, on_step=False, on_epoch=True)
+
+    def on_train_batch_end(self, *args, **kwargs):
+        if self.use_ema:
+            self.model_ema(self.model)
+
+    def _get_rows_from_list(self, samples):
+        n_imgs_per_row = len(samples)
+        denoise_grid = rearrange(samples, 'n b c h w -> b n c h w')
+        denoise_grid = rearrange(denoise_grid, 'b n c h w -> (b n) c h w')
+        denoise_grid = make_grid(denoise_grid, nrow=n_imgs_per_row)
+        return denoise_grid
+
+    @torch.no_grad()
+    def log_images(self, batch, N=8, n_row=2, sample=True, return_keys=None, **kwargs):
+        log = {}
+        x = self.get_input(batch, self.first_stage_key)
+        N = min(x.shape[0], N)
+        n_row = min(x.shape[0], n_row)
+        x = x.to(self.device)[:N]
+        log["inputs"] = x
+
+        # get diffusion row
+        diffusion_row = []
+        x_start = x[:n_row]
+
+        for t in range(self.num_timesteps):
+            if t % self.log_every_t == 0 or t == self.num_timesteps - 1:
+                t = repeat(torch.tensor([t]), '1 -> b', b=n_row)
+                t = t.to(self.device).long()
+                noise = torch.randn_like(x_start)
+                x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
+                diffusion_row.append(x_noisy)
+
+        log["diffusion_row"] = self._get_rows_from_list(diffusion_row)
+
+        if sample:
+            # get denoise row
+            with self.ema_scope("Plotting"):
+                samples, denoise_row = self.sample(batch_size=N, return_intermediates=True)
+
+            log["samples"] = samples
+            log["denoise_row"] = self._get_rows_from_list(denoise_row)
+
+        if return_keys:
+            if np.intersect1d(list(log.keys()), return_keys).shape[0] == 0:
+                return log
+            else:
+                return {key: log[key] for key in return_keys}
+        return log
+
+    def configure_optimizers(self):
+        lr = self.learning_rate
+        params = list(self.model.parameters())
+        if self.learn_logvar:
+            params = params + [self.logvar]
+        opt = torch.optim.AdamW(params, lr=lr)
+        return opt
+
+
+class LatentDiffusionV1(DDPMV1):
+    """main class"""
+    def __init__(self,
+                 first_stage_config,
+                 cond_stage_config,
+                 num_timesteps_cond=None,
+                 cond_stage_key="image",
+                 cond_stage_trainable=False,
+                 concat_mode=True,
+                 cond_stage_forward=None,
+                 conditioning_key=None,
+                 scale_factor=1.0,
+                 scale_by_std=False,
+                 *args, **kwargs):
+        self.num_timesteps_cond = default(num_timesteps_cond, 1)
+        self.scale_by_std = scale_by_std
+        assert self.num_timesteps_cond <= kwargs['timesteps']
+        # for backwards compatibility after implementation of DiffusionWrapper
+        if conditioning_key is None:
+            conditioning_key = 'concat' if concat_mode else 'crossattn'
+        if cond_stage_config == '__is_unconditional__':
+            conditioning_key = None
+        ckpt_path = kwargs.pop("ckpt_path", None)
+        ignore_keys = kwargs.pop("ignore_keys", [])
+        super().__init__(*args, conditioning_key=conditioning_key, **kwargs)
+        self.concat_mode = concat_mode
+        self.cond_stage_trainable = cond_stage_trainable
+        self.cond_stage_key = cond_stage_key
+        try:
+            self.num_downs = len(first_stage_config.params.ddconfig.ch_mult) - 1
+        except Exception:
+            self.num_downs = 0
+        if not scale_by_std:
+            self.scale_factor = scale_factor
+        else:
+            self.register_buffer('scale_factor', torch.tensor(scale_factor))
+        self.instantiate_first_stage(first_stage_config)
+        self.instantiate_cond_stage(cond_stage_config)
+        self.cond_stage_forward = cond_stage_forward
+        self.clip_denoised = False
+        self.bbox_tokenizer = None
+
+        self.restarted_from_ckpt = False
+        if ckpt_path is not None:
+            self.init_from_ckpt(ckpt_path, ignore_keys)
+            self.restarted_from_ckpt = True
+
+    def make_cond_schedule(self, ):
+        self.cond_ids = torch.full(size=(self.num_timesteps,), fill_value=self.num_timesteps - 1, dtype=torch.long)
+        ids = torch.round(torch.linspace(0, self.num_timesteps - 1, self.num_timesteps_cond)).long()
+        self.cond_ids[:self.num_timesteps_cond] = ids
+
+    @rank_zero_only
+    @torch.no_grad()
+    def on_train_batch_start(self, batch, batch_idx, dataloader_idx):
+        # only for very first batch
+        if self.scale_by_std and self.current_epoch == 0 and self.global_step == 0 and batch_idx == 0 and not self.restarted_from_ckpt:
+            assert self.scale_factor == 1., 'rather not use custom rescaling and std-rescaling simultaneously'
+            # set rescale weight to 1./std of encodings
+            print("### USING STD-RESCALING ###")
+            x = super().get_input(batch, self.first_stage_key)
+            x = x.to(self.device)
+            encoder_posterior = self.encode_first_stage(x)
+            z = self.get_first_stage_encoding(encoder_posterior).detach()
+            del self.scale_factor
+            self.register_buffer('scale_factor', 1. / z.flatten().std())
+            print(f"setting self.scale_factor to {self.scale_factor}")
+            print("### USING STD-RESCALING ###")
+
+    def register_schedule(self,
+                          given_betas=None, beta_schedule="linear", timesteps=1000,
+                          linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
+        super().register_schedule(given_betas, beta_schedule, timesteps, linear_start, linear_end, cosine_s)
+
+        self.shorten_cond_schedule = self.num_timesteps_cond > 1
+        if self.shorten_cond_schedule:
+            self.make_cond_schedule()
+
+    def instantiate_first_stage(self, config):
+        model = instantiate_from_config(config)
+        self.first_stage_model = model.eval()
+        self.first_stage_model.train = disabled_train
+        for param in self.first_stage_model.parameters():
+            param.requires_grad = False
+
+    def instantiate_cond_stage(self, config):
+        if not self.cond_stage_trainable:
+            if config == "__is_first_stage__":
+                print("Using first stage also as cond stage.")
+                self.cond_stage_model = self.first_stage_model
+            elif config == "__is_unconditional__":
+                print(f"Training {self.__class__.__name__} as an unconditional model.")
+                self.cond_stage_model = None
+                # self.be_unconditional = True
+            else:
+                model = instantiate_from_config(config)
+                self.cond_stage_model = model.eval()
+                self.cond_stage_model.train = disabled_train
+                for param in self.cond_stage_model.parameters():
+                    param.requires_grad = False
+        else:
+            assert config != '__is_first_stage__'
+            assert config != '__is_unconditional__'
+            model = instantiate_from_config(config)
+            self.cond_stage_model = model
+
+    def _get_denoise_row_from_list(self, samples, desc='', force_no_decoder_quantization=False):
+        denoise_row = []
+        for zd in tqdm(samples, desc=desc):
+            denoise_row.append(self.decode_first_stage(zd.to(self.device),
+                                                            force_not_quantize=force_no_decoder_quantization))
+        n_imgs_per_row = len(denoise_row)
+        denoise_row = torch.stack(denoise_row)  # n_log_step, n_row, C, H, W
+        denoise_grid = rearrange(denoise_row, 'n b c h w -> b n c h w')
+        denoise_grid = rearrange(denoise_grid, 'b n c h w -> (b n) c h w')
+        denoise_grid = make_grid(denoise_grid, nrow=n_imgs_per_row)
+        return denoise_grid
+
+    def get_first_stage_encoding(self, encoder_posterior):
+        if isinstance(encoder_posterior, DiagonalGaussianDistribution):
+            z = encoder_posterior.sample()
+        elif isinstance(encoder_posterior, torch.Tensor):
+            z = encoder_posterior
+        else:
+            raise NotImplementedError(f"encoder_posterior of type '{type(encoder_posterior)}' not yet implemented")
+        return self.scale_factor * z
+
+    def get_learned_conditioning(self, c):
+        if self.cond_stage_forward is None:
+            if hasattr(self.cond_stage_model, 'encode') and callable(self.cond_stage_model.encode):
+                c = self.cond_stage_model.encode(c)
+                if isinstance(c, DiagonalGaussianDistribution):
+                    c = c.mode()
+            else:
+                c = self.cond_stage_model(c)
+        else:
+            assert hasattr(self.cond_stage_model, self.cond_stage_forward)
+            c = getattr(self.cond_stage_model, self.cond_stage_forward)(c)
+        return c
+
+    def meshgrid(self, h, w):
+        y = torch.arange(0, h).view(h, 1, 1).repeat(1, w, 1)
+        x = torch.arange(0, w).view(1, w, 1).repeat(h, 1, 1)
+
+        arr = torch.cat([y, x], dim=-1)
+        return arr
+
+    def delta_border(self, h, w):
+        """
+        :param h: height
+        :param w: width
+        :return: normalized distance to image border,
+         wtith min distance = 0 at border and max dist = 0.5 at image center
+        """
+        lower_right_corner = torch.tensor([h - 1, w - 1]).view(1, 1, 2)
+        arr = self.meshgrid(h, w) / lower_right_corner
+        dist_left_up = torch.min(arr, dim=-1, keepdims=True)[0]
+        dist_right_down = torch.min(1 - arr, dim=-1, keepdims=True)[0]
+        edge_dist = torch.min(torch.cat([dist_left_up, dist_right_down], dim=-1), dim=-1)[0]
+        return edge_dist
+
+    def get_weighting(self, h, w, Ly, Lx, device):
+        weighting = self.delta_border(h, w)
+        weighting = torch.clip(weighting, self.split_input_params["clip_min_weight"],
+                               self.split_input_params["clip_max_weight"], )
+        weighting = weighting.view(1, h * w, 1).repeat(1, 1, Ly * Lx).to(device)
+
+        if self.split_input_params["tie_braker"]:
+            L_weighting = self.delta_border(Ly, Lx)
+            L_weighting = torch.clip(L_weighting,
+                                     self.split_input_params["clip_min_tie_weight"],
+                                     self.split_input_params["clip_max_tie_weight"])
+
+            L_weighting = L_weighting.view(1, 1, Ly * Lx).to(device)
+            weighting = weighting * L_weighting
+        return weighting
+
+    def get_fold_unfold(self, x, kernel_size, stride, uf=1, df=1):  # todo load once not every time, shorten code
+        """
+        :param x: img of size (bs, c, h, w)
+        :return: n img crops of size (n, bs, c, kernel_size[0], kernel_size[1])
+        """
+        bs, nc, h, w = x.shape
+
+        # number of crops in image
+        Ly = (h - kernel_size[0]) // stride[0] + 1
+        Lx = (w - kernel_size[1]) // stride[1] + 1
+
+        if uf == 1 and df == 1:
+            fold_params = dict(kernel_size=kernel_size, dilation=1, padding=0, stride=stride)
+            unfold = torch.nn.Unfold(**fold_params)
+
+            fold = torch.nn.Fold(output_size=x.shape[2:], **fold_params)
+
+            weighting = self.get_weighting(kernel_size[0], kernel_size[1], Ly, Lx, x.device).to(x.dtype)
+            normalization = fold(weighting).view(1, 1, h, w)  # normalizes the overlap
+            weighting = weighting.view((1, 1, kernel_size[0], kernel_size[1], Ly * Lx))
+
+        elif uf > 1 and df == 1:
+            fold_params = dict(kernel_size=kernel_size, dilation=1, padding=0, stride=stride)
+            unfold = torch.nn.Unfold(**fold_params)
+
+            fold_params2 = dict(kernel_size=(kernel_size[0] * uf, kernel_size[0] * uf),
+                                dilation=1, padding=0,
+                                stride=(stride[0] * uf, stride[1] * uf))
+            fold = torch.nn.Fold(output_size=(x.shape[2] * uf, x.shape[3] * uf), **fold_params2)
+
+            weighting = self.get_weighting(kernel_size[0] * uf, kernel_size[1] * uf, Ly, Lx, x.device).to(x.dtype)
+            normalization = fold(weighting).view(1, 1, h * uf, w * uf)  # normalizes the overlap
+            weighting = weighting.view((1, 1, kernel_size[0] * uf, kernel_size[1] * uf, Ly * Lx))
+
+        elif df > 1 and uf == 1:
+            fold_params = dict(kernel_size=kernel_size, dilation=1, padding=0, stride=stride)
+            unfold = torch.nn.Unfold(**fold_params)
+
+            fold_params2 = dict(kernel_size=(kernel_size[0] // df, kernel_size[0] // df),
+                                dilation=1, padding=0,
+                                stride=(stride[0] // df, stride[1] // df))
+            fold = torch.nn.Fold(output_size=(x.shape[2] // df, x.shape[3] // df), **fold_params2)
+
+            weighting = self.get_weighting(kernel_size[0] // df, kernel_size[1] // df, Ly, Lx, x.device).to(x.dtype)
+            normalization = fold(weighting).view(1, 1, h // df, w // df)  # normalizes the overlap
+            weighting = weighting.view((1, 1, kernel_size[0] // df, kernel_size[1] // df, Ly * Lx))
+
+        else:
+            raise NotImplementedError
+
+        return fold, unfold, normalization, weighting
+
+    @torch.no_grad()
+    def get_input(self, batch, k, return_first_stage_outputs=False, force_c_encode=False,
+                  cond_key=None, return_original_cond=False, bs=None):
+        x = super().get_input(batch, k)
+        if bs is not None:
+            x = x[:bs]
+        x = x.to(self.device)
+        encoder_posterior = self.encode_first_stage(x)
+        z = self.get_first_stage_encoding(encoder_posterior).detach()
+
+        if self.model.conditioning_key is not None:
+            if cond_key is None:
+                cond_key = self.cond_stage_key
+            if cond_key != self.first_stage_key:
+                if cond_key in ['caption', 'coordinates_bbox']:
+                    xc = batch[cond_key]
+                elif cond_key == 'class_label':
+                    xc = batch
+                else:
+                    xc = super().get_input(batch, cond_key).to(self.device)
+            else:
+                xc = x
+            if not self.cond_stage_trainable or force_c_encode:
+                if isinstance(xc, dict) or isinstance(xc, list):
+                    # import pudb; pudb.set_trace()
+                    c = self.get_learned_conditioning(xc)
+                else:
+                    c = self.get_learned_conditioning(xc.to(self.device))
+            else:
+                c = xc
+            if bs is not None:
+                c = c[:bs]
+
+            if self.use_positional_encodings:
+                pos_x, pos_y = self.compute_latent_shifts(batch)
+                ckey = __conditioning_keys__[self.model.conditioning_key]
+                c = {ckey: c, 'pos_x': pos_x, 'pos_y': pos_y}
+
+        else:
+            c = None
+            xc = None
+            if self.use_positional_encodings:
+                pos_x, pos_y = self.compute_latent_shifts(batch)
+                c = {'pos_x': pos_x, 'pos_y': pos_y}
+        out = [z, c]
+        if return_first_stage_outputs:
+            xrec = self.decode_first_stage(z)
+            out.extend([x, xrec])
+        if return_original_cond:
+            out.append(xc)
+        return out
+
+    @torch.no_grad()
+    def decode_first_stage(self, z, predict_cids=False, force_not_quantize=False):
+        if predict_cids:
+            if z.dim() == 4:
+                z = torch.argmax(z.exp(), dim=1).long()
+            z = self.first_stage_model.quantize.get_codebook_entry(z, shape=None)
+            z = rearrange(z, 'b h w c -> b c h w').contiguous()
+
+        z = 1. / self.scale_factor * z
+
+        if hasattr(self, "split_input_params"):
+            if self.split_input_params["patch_distributed_vq"]:
+                ks = self.split_input_params["ks"]  # eg. (128, 128)
+                stride = self.split_input_params["stride"]  # eg. (64, 64)
+                uf = self.split_input_params["vqf"]
+                bs, nc, h, w = z.shape
+                if ks[0] > h or ks[1] > w:
+                    ks = (min(ks[0], h), min(ks[1], w))
+                    print("reducing Kernel")
+
+                if stride[0] > h or stride[1] > w:
+                    stride = (min(stride[0], h), min(stride[1], w))
+                    print("reducing stride")
+
+                fold, unfold, normalization, weighting = self.get_fold_unfold(z, ks, stride, uf=uf)
+
+                z = unfold(z)  # (bn, nc * prod(**ks), L)
+                # 1. Reshape to img shape
+                z = z.view((z.shape[0], -1, ks[0], ks[1], z.shape[-1]))  # (bn, nc, ks[0], ks[1], L )
+
+                # 2. apply model loop over last dim
+                if isinstance(self.first_stage_model, VQModelInterface):
+                    output_list = [self.first_stage_model.decode(z[:, :, :, :, i],
+                                                                 force_not_quantize=predict_cids or force_not_quantize)
+                                   for i in range(z.shape[-1])]
+                else:
+
+                    output_list = [self.first_stage_model.decode(z[:, :, :, :, i])
+                                   for i in range(z.shape[-1])]
+
+                o = torch.stack(output_list, axis=-1)  # # (bn, nc, ks[0], ks[1], L)
+                o = o * weighting
+                # Reverse 1. reshape to img shape
+                o = o.view((o.shape[0], -1, o.shape[-1]))  # (bn, nc * ks[0] * ks[1], L)
+                # stitch crops together
+                decoded = fold(o)
+                decoded = decoded / normalization  # norm is shape (1, 1, h, w)
+                return decoded
+            else:
+                if isinstance(self.first_stage_model, VQModelInterface):
+                    return self.first_stage_model.decode(z, force_not_quantize=predict_cids or force_not_quantize)
+                else:
+                    return self.first_stage_model.decode(z)
+
+        else:
+            if isinstance(self.first_stage_model, VQModelInterface):
+                return self.first_stage_model.decode(z, force_not_quantize=predict_cids or force_not_quantize)
+            else:
+                return self.first_stage_model.decode(z)
+
+    # same as above but without decorator
+    def differentiable_decode_first_stage(self, z, predict_cids=False, force_not_quantize=False):
+        if predict_cids:
+            if z.dim() == 4:
+                z = torch.argmax(z.exp(), dim=1).long()
+            z = self.first_stage_model.quantize.get_codebook_entry(z, shape=None)
+            z = rearrange(z, 'b h w c -> b c h w').contiguous()
+
+        z = 1. / self.scale_factor * z
+
+        if hasattr(self, "split_input_params"):
+            if self.split_input_params["patch_distributed_vq"]:
+                ks = self.split_input_params["ks"]  # eg. (128, 128)
+                stride = self.split_input_params["stride"]  # eg. (64, 64)
+                uf = self.split_input_params["vqf"]
+                bs, nc, h, w = z.shape
+                if ks[0] > h or ks[1] > w:
+                    ks = (min(ks[0], h), min(ks[1], w))
+                    print("reducing Kernel")
+
+                if stride[0] > h or stride[1] > w:
+                    stride = (min(stride[0], h), min(stride[1], w))
+                    print("reducing stride")
+
+                fold, unfold, normalization, weighting = self.get_fold_unfold(z, ks, stride, uf=uf)
+
+                z = unfold(z)  # (bn, nc * prod(**ks), L)
+                # 1. Reshape to img shape
+                z = z.view((z.shape[0], -1, ks[0], ks[1], z.shape[-1]))  # (bn, nc, ks[0], ks[1], L )
+
+                # 2. apply model loop over last dim
+                if isinstance(self.first_stage_model, VQModelInterface):
+                    output_list = [self.first_stage_model.decode(z[:, :, :, :, i],
+                                                                 force_not_quantize=predict_cids or force_not_quantize)
+                                   for i in range(z.shape[-1])]
+                else:
+
+                    output_list = [self.first_stage_model.decode(z[:, :, :, :, i])
+                                   for i in range(z.shape[-1])]
+
+                o = torch.stack(output_list, axis=-1)  # # (bn, nc, ks[0], ks[1], L)
+                o = o * weighting
+                # Reverse 1. reshape to img shape
+                o = o.view((o.shape[0], -1, o.shape[-1]))  # (bn, nc * ks[0] * ks[1], L)
+                # stitch crops together
+                decoded = fold(o)
+                decoded = decoded / normalization  # norm is shape (1, 1, h, w)
+                return decoded
+            else:
+                if isinstance(self.first_stage_model, VQModelInterface):
+                    return self.first_stage_model.decode(z, force_not_quantize=predict_cids or force_not_quantize)
+                else:
+                    return self.first_stage_model.decode(z)
+
+        else:
+            if isinstance(self.first_stage_model, VQModelInterface):
+                return self.first_stage_model.decode(z, force_not_quantize=predict_cids or force_not_quantize)
+            else:
+                return self.first_stage_model.decode(z)
+
+    @torch.no_grad()
+    def encode_first_stage(self, x):
+        if hasattr(self, "split_input_params"):
+            if self.split_input_params["patch_distributed_vq"]:
+                ks = self.split_input_params["ks"]  # eg. (128, 128)
+                stride = self.split_input_params["stride"]  # eg. (64, 64)
+                df = self.split_input_params["vqf"]
+                self.split_input_params['original_image_size'] = x.shape[-2:]
+                bs, nc, h, w = x.shape
+                if ks[0] > h or ks[1] > w:
+                    ks = (min(ks[0], h), min(ks[1], w))
+                    print("reducing Kernel")
+
+                if stride[0] > h or stride[1] > w:
+                    stride = (min(stride[0], h), min(stride[1], w))
+                    print("reducing stride")
+
+                fold, unfold, normalization, weighting = self.get_fold_unfold(x, ks, stride, df=df)
+                z = unfold(x)  # (bn, nc * prod(**ks), L)
+                # Reshape to img shape
+                z = z.view((z.shape[0], -1, ks[0], ks[1], z.shape[-1]))  # (bn, nc, ks[0], ks[1], L )
+
+                output_list = [self.first_stage_model.encode(z[:, :, :, :, i])
+                               for i in range(z.shape[-1])]
+
+                o = torch.stack(output_list, axis=-1)
+                o = o * weighting
+
+                # Reverse reshape to img shape
+                o = o.view((o.shape[0], -1, o.shape[-1]))  # (bn, nc * ks[0] * ks[1], L)
+                # stitch crops together
+                decoded = fold(o)
+                decoded = decoded / normalization
+                return decoded
+
+            else:
+                return self.first_stage_model.encode(x)
+        else:
+            return self.first_stage_model.encode(x)
+
+    def shared_step(self, batch, **kwargs):
+        x, c = self.get_input(batch, self.first_stage_key)
+        loss = self(x, c)
+        return loss
+
+    def forward(self, x, c, *args, **kwargs):
+        t = torch.randint(0, self.num_timesteps, (x.shape[0],), device=self.device).long()
+        if self.model.conditioning_key is not None:
+            assert c is not None
+            if self.cond_stage_trainable:
+                c = self.get_learned_conditioning(c)
+            if self.shorten_cond_schedule:  # TODO: drop this option
+                tc = self.cond_ids[t].to(self.device)
+                c = self.q_sample(x_start=c, t=tc, noise=torch.randn_like(c.float()))
+        return self.p_losses(x, c, t, *args, **kwargs)
+
+    def apply_model(self, x_noisy, t, cond, return_ids=False):
+
+        if isinstance(cond, dict):
+            # hybrid case, cond is exptected to be a dict
+            pass
+        else:
+            if not isinstance(cond, list):
+                cond = [cond]
+            key = 'c_concat' if self.model.conditioning_key == 'concat' else 'c_crossattn'
+            cond = {key: cond}
+
+        if hasattr(self, "split_input_params"):
+            assert len(cond) == 1  # todo can only deal with one conditioning atm
+            assert not return_ids
+            ks = self.split_input_params["ks"]  # eg. (128, 128)
+            stride = self.split_input_params["stride"]  # eg. (64, 64)
+
+            h, w = x_noisy.shape[-2:]
+
+            fold, unfold, normalization, weighting = self.get_fold_unfold(x_noisy, ks, stride)
+
+            z = unfold(x_noisy)  # (bn, nc * prod(**ks), L)
+            # Reshape to img shape
+            z = z.view((z.shape[0], -1, ks[0], ks[1], z.shape[-1]))  # (bn, nc, ks[0], ks[1], L )
+            z_list = [z[:, :, :, :, i] for i in range(z.shape[-1])]
+
+            if self.cond_stage_key in ["image", "LR_image", "segmentation",
+                                       'bbox_img'] and self.model.conditioning_key:  # todo check for completeness
+                c_key = next(iter(cond.keys()))  # get key
+                c = next(iter(cond.values()))  # get value
+                assert (len(c) == 1)  # todo extend to list with more than one elem
+                c = c[0]  # get element
+
+                c = unfold(c)
+                c = c.view((c.shape[0], -1, ks[0], ks[1], c.shape[-1]))  # (bn, nc, ks[0], ks[1], L )
+
+                cond_list = [{c_key: [c[:, :, :, :, i]]} for i in range(c.shape[-1])]
+
+            elif self.cond_stage_key == 'coordinates_bbox':
+                assert 'original_image_size' in self.split_input_params, 'BoudingBoxRescaling is missing original_image_size'
+
+                # assuming padding of unfold is always 0 and its dilation is always 1
+                n_patches_per_row = int((w - ks[0]) / stride[0] + 1)
+                full_img_h, full_img_w = self.split_input_params['original_image_size']
+                # as we are operating on latents, we need the factor from the original image size to the
+                # spatial latent size to properly rescale the crops for regenerating the bbox annotations
+                num_downs = self.first_stage_model.encoder.num_resolutions - 1
+                rescale_latent = 2 ** (num_downs)
+
+                # get top left postions of patches as conforming for the bbbox tokenizer, therefore we
+                # need to rescale the tl patch coordinates to be in between (0,1)
+                tl_patch_coordinates = [(rescale_latent * stride[0] * (patch_nr % n_patches_per_row) / full_img_w,
+                                         rescale_latent * stride[1] * (patch_nr // n_patches_per_row) / full_img_h)
+                                        for patch_nr in range(z.shape[-1])]
+
+                # patch_limits are tl_coord, width and height coordinates as (x_tl, y_tl, h, w)
+                patch_limits = [(x_tl, y_tl,
+                                 rescale_latent * ks[0] / full_img_w,
+                                 rescale_latent * ks[1] / full_img_h) for x_tl, y_tl in tl_patch_coordinates]
+                # patch_values = [(np.arange(x_tl,min(x_tl+ks, 1.)),np.arange(y_tl,min(y_tl+ks, 1.))) for x_tl, y_tl in tl_patch_coordinates]
+
+                # tokenize crop coordinates for the bounding boxes of the respective patches
+                patch_limits_tknzd = [torch.LongTensor(self.bbox_tokenizer._crop_encoder(bbox))[None].to(self.device)
+                                      for bbox in patch_limits]  # list of length l with tensors of shape (1, 2)
+                print(patch_limits_tknzd[0].shape)
+                # cut tknzd crop position from conditioning
+                assert isinstance(cond, dict), 'cond must be dict to be fed into model'
+                cut_cond = cond['c_crossattn'][0][..., :-2].to(self.device)
+                print(cut_cond.shape)
+
+                adapted_cond = torch.stack([torch.cat([cut_cond, p], dim=1) for p in patch_limits_tknzd])
+                adapted_cond = rearrange(adapted_cond, 'l b n -> (l b) n')
+                print(adapted_cond.shape)
+                adapted_cond = self.get_learned_conditioning(adapted_cond)
+                print(adapted_cond.shape)
+                adapted_cond = rearrange(adapted_cond, '(l b) n d -> l b n d', l=z.shape[-1])
+                print(adapted_cond.shape)
+
+                cond_list = [{'c_crossattn': [e]} for e in adapted_cond]
+
+            else:
+                cond_list = [cond for i in range(z.shape[-1])]  # Todo make this more efficient
+
+            # apply model by loop over crops
+            output_list = [self.model(z_list[i], t, **cond_list[i]) for i in range(z.shape[-1])]
+            assert not isinstance(output_list[0],
+                                  tuple)  # todo cant deal with multiple model outputs check this never happens
+
+            o = torch.stack(output_list, axis=-1)
+            o = o * weighting
+            # Reverse reshape to img shape
+            o = o.view((o.shape[0], -1, o.shape[-1]))  # (bn, nc * ks[0] * ks[1], L)
+            # stitch crops together
+            x_recon = fold(o) / normalization
+
+        else:
+            x_recon = self.model(x_noisy, t, **cond)
+
+        if isinstance(x_recon, tuple) and not return_ids:
+            return x_recon[0]
+        else:
+            return x_recon
+
+    def _predict_eps_from_xstart(self, x_t, t, pred_xstart):
+        return (extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t - pred_xstart) / \
+               extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape)
+
+    def _prior_bpd(self, x_start):
+        """
+        Get the prior KL term for the variational lower-bound, measured in
+        bits-per-dim.
+        This term can't be optimized, as it only depends on the encoder.
+        :param x_start: the [N x C x ...] tensor of inputs.
+        :return: a batch of [N] KL values (in bits), one per batch element.
+        """
+        batch_size = x_start.shape[0]
+        t = torch.tensor([self.num_timesteps - 1] * batch_size, device=x_start.device)
+        qt_mean, _, qt_log_variance = self.q_mean_variance(x_start, t)
+        kl_prior = normal_kl(mean1=qt_mean, logvar1=qt_log_variance, mean2=0.0, logvar2=0.0)
+        return mean_flat(kl_prior) / np.log(2.0)
+
+    def p_losses(self, x_start, cond, t, noise=None):
+        noise = default(noise, lambda: torch.randn_like(x_start))
+        x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
+        model_output = self.apply_model(x_noisy, t, cond)
+
+        loss_dict = {}
+        prefix = 'train' if self.training else 'val'
+
+        if self.parameterization == "x0":
+            target = x_start
+        elif self.parameterization == "eps":
+            target = noise
+        else:
+            raise NotImplementedError()
+
+        loss_simple = self.get_loss(model_output, target, mean=False).mean([1, 2, 3])
+        loss_dict.update({f'{prefix}/loss_simple': loss_simple.mean()})
+
+        logvar_t = self.logvar[t].to(self.device)
+        loss = loss_simple / torch.exp(logvar_t) + logvar_t
+        # loss = loss_simple / torch.exp(self.logvar) + self.logvar
+        if self.learn_logvar:
+            loss_dict.update({f'{prefix}/loss_gamma': loss.mean()})
+            loss_dict.update({'logvar': self.logvar.data.mean()})
+
+        loss = self.l_simple_weight * loss.mean()
+
+        loss_vlb = self.get_loss(model_output, target, mean=False).mean(dim=(1, 2, 3))
+        loss_vlb = (self.lvlb_weights[t] * loss_vlb).mean()
+        loss_dict.update({f'{prefix}/loss_vlb': loss_vlb})
+        loss += (self.original_elbo_weight * loss_vlb)
+        loss_dict.update({f'{prefix}/loss': loss})
+
+        return loss, loss_dict
+
+    def p_mean_variance(self, x, c, t, clip_denoised: bool, return_codebook_ids=False, quantize_denoised=False,
+                        return_x0=False, score_corrector=None, corrector_kwargs=None):
+        t_in = t
+        model_out = self.apply_model(x, t_in, c, return_ids=return_codebook_ids)
+
+        if score_corrector is not None:
+            assert self.parameterization == "eps"
+            model_out = score_corrector.modify_score(self, model_out, x, t, c, **corrector_kwargs)
+
+        if return_codebook_ids:
+            model_out, logits = model_out
+
+        if self.parameterization == "eps":
+            x_recon = self.predict_start_from_noise(x, t=t, noise=model_out)
+        elif self.parameterization == "x0":
+            x_recon = model_out
+        else:
+            raise NotImplementedError()
+
+        if clip_denoised:
+            x_recon.clamp_(-1., 1.)
+        if quantize_denoised:
+            x_recon, _, [_, _, indices] = self.first_stage_model.quantize(x_recon)
+        model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
+        if return_codebook_ids:
+            return model_mean, posterior_variance, posterior_log_variance, logits
+        elif return_x0:
+            return model_mean, posterior_variance, posterior_log_variance, x_recon
+        else:
+            return model_mean, posterior_variance, posterior_log_variance
+
+    @torch.no_grad()
+    def p_sample(self, x, c, t, clip_denoised=False, repeat_noise=False,
+                 return_codebook_ids=False, quantize_denoised=False, return_x0=False,
+                 temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None):
+        b, *_, device = *x.shape, x.device
+        outputs = self.p_mean_variance(x=x, c=c, t=t, clip_denoised=clip_denoised,
+                                       return_codebook_ids=return_codebook_ids,
+                                       quantize_denoised=quantize_denoised,
+                                       return_x0=return_x0,
+                                       score_corrector=score_corrector, corrector_kwargs=corrector_kwargs)
+        if return_codebook_ids:
+            raise DeprecationWarning("Support dropped.")
+            model_mean, _, model_log_variance, logits = outputs
+        elif return_x0:
+            model_mean, _, model_log_variance, x0 = outputs
+        else:
+            model_mean, _, model_log_variance = outputs
+
+        noise = noise_like(x.shape, device, repeat_noise) * temperature
+        if noise_dropout > 0.:
+            noise = torch.nn.functional.dropout(noise, p=noise_dropout)
+        # no noise when t == 0
+        nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
+
+        if return_codebook_ids:
+            return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise, logits.argmax(dim=1)
+        if return_x0:
+            return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise, x0
+        else:
+            return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
+
+    @torch.no_grad()
+    def progressive_denoising(self, cond, shape, verbose=True, callback=None, quantize_denoised=False,
+                              img_callback=None, mask=None, x0=None, temperature=1., noise_dropout=0.,
+                              score_corrector=None, corrector_kwargs=None, batch_size=None, x_T=None, start_T=None,
+                              log_every_t=None):
+        if not log_every_t:
+            log_every_t = self.log_every_t
+        timesteps = self.num_timesteps
+        if batch_size is not None:
+            b = batch_size if batch_size is not None else shape[0]
+            shape = [batch_size] + list(shape)
+        else:
+            b = batch_size = shape[0]
+        if x_T is None:
+            img = torch.randn(shape, device=self.device)
+        else:
+            img = x_T
+        intermediates = []
+        if cond is not None:
+            if isinstance(cond, dict):
+                cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else
+                [x[:batch_size] for x in cond[key]] for key in cond}
+            else:
+                cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size]
+
+        if start_T is not None:
+            timesteps = min(timesteps, start_T)
+        iterator = tqdm(reversed(range(0, timesteps)), desc='Progressive Generation',
+                        total=timesteps) if verbose else reversed(
+            range(0, timesteps))
+        if type(temperature) == float:
+            temperature = [temperature] * timesteps
+
+        for i in iterator:
+            ts = torch.full((b,), i, device=self.device, dtype=torch.long)
+            if self.shorten_cond_schedule:
+                assert self.model.conditioning_key != 'hybrid'
+                tc = self.cond_ids[ts].to(cond.device)
+                cond = self.q_sample(x_start=cond, t=tc, noise=torch.randn_like(cond))
+
+            img, x0_partial = self.p_sample(img, cond, ts,
+                                            clip_denoised=self.clip_denoised,
+                                            quantize_denoised=quantize_denoised, return_x0=True,
+                                            temperature=temperature[i], noise_dropout=noise_dropout,
+                                            score_corrector=score_corrector, corrector_kwargs=corrector_kwargs)
+            if mask is not None:
+                assert x0 is not None
+                img_orig = self.q_sample(x0, ts)
+                img = img_orig * mask + (1. - mask) * img
+
+            if i % log_every_t == 0 or i == timesteps - 1:
+                intermediates.append(x0_partial)
+            if callback:
+                callback(i)
+            if img_callback:
+                img_callback(img, i)
+        return img, intermediates
+
+    @torch.no_grad()
+    def p_sample_loop(self, cond, shape, return_intermediates=False,
+                      x_T=None, verbose=True, callback=None, timesteps=None, quantize_denoised=False,
+                      mask=None, x0=None, img_callback=None, start_T=None,
+                      log_every_t=None):
+
+        if not log_every_t:
+            log_every_t = self.log_every_t
+        device = self.betas.device
+        b = shape[0]
+        if x_T is None:
+            img = torch.randn(shape, device=device)
+        else:
+            img = x_T
+
+        intermediates = [img]
+        if timesteps is None:
+            timesteps = self.num_timesteps
+
+        if start_T is not None:
+            timesteps = min(timesteps, start_T)
+        iterator = tqdm(reversed(range(0, timesteps)), desc='Sampling t', total=timesteps) if verbose else reversed(
+            range(0, timesteps))
+
+        if mask is not None:
+            assert x0 is not None
+            assert x0.shape[2:3] == mask.shape[2:3]  # spatial size has to match
+
+        for i in iterator:
+            ts = torch.full((b,), i, device=device, dtype=torch.long)
+            if self.shorten_cond_schedule:
+                assert self.model.conditioning_key != 'hybrid'
+                tc = self.cond_ids[ts].to(cond.device)
+                cond = self.q_sample(x_start=cond, t=tc, noise=torch.randn_like(cond))
+
+            img = self.p_sample(img, cond, ts,
+                                clip_denoised=self.clip_denoised,
+                                quantize_denoised=quantize_denoised)
+            if mask is not None:
+                img_orig = self.q_sample(x0, ts)
+                img = img_orig * mask + (1. - mask) * img
+
+            if i % log_every_t == 0 or i == timesteps - 1:
+                intermediates.append(img)
+            if callback:
+                callback(i)
+            if img_callback:
+                img_callback(img, i)
+
+        if return_intermediates:
+            return img, intermediates
+        return img
+
+    @torch.no_grad()
+    def sample(self, cond, batch_size=16, return_intermediates=False, x_T=None,
+               verbose=True, timesteps=None, quantize_denoised=False,
+               mask=None, x0=None, shape=None,**kwargs):
+        if shape is None:
+            shape = (batch_size, self.channels, self.image_size, self.image_size)
+        if cond is not None:
+            if isinstance(cond, dict):
+                cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else
+                [x[:batch_size] for x in cond[key]] for key in cond}
+            else:
+                cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size]
+        return self.p_sample_loop(cond,
+                                  shape,
+                                  return_intermediates=return_intermediates, x_T=x_T,
+                                  verbose=verbose, timesteps=timesteps, quantize_denoised=quantize_denoised,
+                                  mask=mask, x0=x0)
+
+    @torch.no_grad()
+    def sample_log(self,cond,batch_size,ddim, ddim_steps,**kwargs):
+
+        if ddim:
+            ddim_sampler = DDIMSampler(self)
+            shape = (self.channels, self.image_size, self.image_size)
+            samples, intermediates =ddim_sampler.sample(ddim_steps,batch_size,
+                                                        shape,cond,verbose=False,**kwargs)
+
+        else:
+            samples, intermediates = self.sample(cond=cond, batch_size=batch_size,
+                                                 return_intermediates=True,**kwargs)
+
+        return samples, intermediates
+
+
+    @torch.no_grad()
+    def log_images(self, batch, N=8, n_row=4, sample=True, ddim_steps=200, ddim_eta=1., return_keys=None,
+                   quantize_denoised=True, inpaint=True, plot_denoise_rows=False, plot_progressive_rows=True,
+                   plot_diffusion_rows=True, **kwargs):
+
+        use_ddim = ddim_steps is not None
+
+        log = {}
+        z, c, x, xrec, xc = self.get_input(batch, self.first_stage_key,
+                                           return_first_stage_outputs=True,
+                                           force_c_encode=True,
+                                           return_original_cond=True,
+                                           bs=N)
+        N = min(x.shape[0], N)
+        n_row = min(x.shape[0], n_row)
+        log["inputs"] = x
+        log["reconstruction"] = xrec
+        if self.model.conditioning_key is not None:
+            if hasattr(self.cond_stage_model, "decode"):
+                xc = self.cond_stage_model.decode(c)
+                log["conditioning"] = xc
+            elif self.cond_stage_key in ["caption"]:
+                xc = log_txt_as_img((x.shape[2], x.shape[3]), batch["caption"])
+                log["conditioning"] = xc
+            elif self.cond_stage_key == 'class_label':
+                xc = log_txt_as_img((x.shape[2], x.shape[3]), batch["human_label"])
+                log['conditioning'] = xc
+            elif isimage(xc):
+                log["conditioning"] = xc
+            if ismap(xc):
+                log["original_conditioning"] = self.to_rgb(xc)
+
+        if plot_diffusion_rows:
+            # get diffusion row
+            diffusion_row = []
+            z_start = z[:n_row]
+            for t in range(self.num_timesteps):
+                if t % self.log_every_t == 0 or t == self.num_timesteps - 1:
+                    t = repeat(torch.tensor([t]), '1 -> b', b=n_row)
+                    t = t.to(self.device).long()
+                    noise = torch.randn_like(z_start)
+                    z_noisy = self.q_sample(x_start=z_start, t=t, noise=noise)
+                    diffusion_row.append(self.decode_first_stage(z_noisy))
+
+            diffusion_row = torch.stack(diffusion_row)  # n_log_step, n_row, C, H, W
+            diffusion_grid = rearrange(diffusion_row, 'n b c h w -> b n c h w')
+            diffusion_grid = rearrange(diffusion_grid, 'b n c h w -> (b n) c h w')
+            diffusion_grid = make_grid(diffusion_grid, nrow=diffusion_row.shape[0])
+            log["diffusion_row"] = diffusion_grid
+
+        if sample:
+            # get denoise row
+            with self.ema_scope("Plotting"):
+                samples, z_denoise_row = self.sample_log(cond=c,batch_size=N,ddim=use_ddim,
+                                                         ddim_steps=ddim_steps,eta=ddim_eta)
+                # samples, z_denoise_row = self.sample(cond=c, batch_size=N, return_intermediates=True)
+            x_samples = self.decode_first_stage(samples)
+            log["samples"] = x_samples
+            if plot_denoise_rows:
+                denoise_grid = self._get_denoise_row_from_list(z_denoise_row)
+                log["denoise_row"] = denoise_grid
+
+            if quantize_denoised and not isinstance(self.first_stage_model, AutoencoderKL) and not isinstance(
+                    self.first_stage_model, IdentityFirstStage):
+                # also display when quantizing x0 while sampling
+                with self.ema_scope("Plotting Quantized Denoised"):
+                    samples, z_denoise_row = self.sample_log(cond=c,batch_size=N,ddim=use_ddim,
+                                                             ddim_steps=ddim_steps,eta=ddim_eta,
+                                                             quantize_denoised=True)
+                    # samples, z_denoise_row = self.sample(cond=c, batch_size=N, return_intermediates=True,
+                    #                                      quantize_denoised=True)
+                x_samples = self.decode_first_stage(samples.to(self.device))
+                log["samples_x0_quantized"] = x_samples
+
+            if inpaint:
+                # make a simple center square
+                h, w = z.shape[2], z.shape[3]
+                mask = torch.ones(N, h, w).to(self.device)
+                # zeros will be filled in
+                mask[:, h // 4:3 * h // 4, w // 4:3 * w // 4] = 0.
+                mask = mask[:, None, ...]
+                with self.ema_scope("Plotting Inpaint"):
+
+                    samples, _ = self.sample_log(cond=c,batch_size=N,ddim=use_ddim, eta=ddim_eta,
+                                                ddim_steps=ddim_steps, x0=z[:N], mask=mask)
+                x_samples = self.decode_first_stage(samples.to(self.device))
+                log["samples_inpainting"] = x_samples
+                log["mask"] = mask
+
+                # outpaint
+                with self.ema_scope("Plotting Outpaint"):
+                    samples, _ = self.sample_log(cond=c, batch_size=N, ddim=use_ddim,eta=ddim_eta,
+                                                ddim_steps=ddim_steps, x0=z[:N], mask=mask)
+                x_samples = self.decode_first_stage(samples.to(self.device))
+                log["samples_outpainting"] = x_samples
+
+        if plot_progressive_rows:
+            with self.ema_scope("Plotting Progressives"):
+                img, progressives = self.progressive_denoising(c,
+                                                               shape=(self.channels, self.image_size, self.image_size),
+                                                               batch_size=N)
+            prog_row = self._get_denoise_row_from_list(progressives, desc="Progressive Generation")
+            log["progressive_row"] = prog_row
+
+        if return_keys:
+            if np.intersect1d(list(log.keys()), return_keys).shape[0] == 0:
+                return log
+            else:
+                return {key: log[key] for key in return_keys}
+        return log
+
+    def configure_optimizers(self):
+        lr = self.learning_rate
+        params = list(self.model.parameters())
+        if self.cond_stage_trainable:
+            print(f"{self.__class__.__name__}: Also optimizing conditioner params!")
+            params = params + list(self.cond_stage_model.parameters())
+        if self.learn_logvar:
+            print('Diffusion model optimizing logvar')
+            params.append(self.logvar)
+        opt = torch.optim.AdamW(params, lr=lr)
+        if self.use_scheduler:
+            assert 'target' in self.scheduler_config
+            scheduler = instantiate_from_config(self.scheduler_config)
+
+            print("Setting up LambdaLR scheduler...")
+            scheduler = [
+                {
+                    'scheduler': LambdaLR(opt, lr_lambda=scheduler.schedule),
+                    'interval': 'step',
+                    'frequency': 1
+                }]
+            return [opt], scheduler
+        return opt
+
+    @torch.no_grad()
+    def to_rgb(self, x):
+        x = x.float()
+        if not hasattr(self, "colorize"):
+            self.colorize = torch.randn(3, x.shape[1], 1, 1).to(x)
+        x = nn.functional.conv2d(x, weight=self.colorize)
+        x = 2. * (x - x.min()) / (x.max() - x.min()) - 1.
+        return x
+
+
+class DiffusionWrapperV1(pl.LightningModule):
+    def __init__(self, diff_model_config, conditioning_key):
+        super().__init__()
+        self.diffusion_model = instantiate_from_config(diff_model_config)
+        self.conditioning_key = conditioning_key
+        assert self.conditioning_key in [None, 'concat', 'crossattn', 'hybrid', 'adm']
+
+    def forward(self, x, t, c_concat: list = None, c_crossattn: list = None):
+        if self.conditioning_key is None:
+            out = self.diffusion_model(x, t)
+        elif self.conditioning_key == 'concat':
+            xc = torch.cat([x] + c_concat, dim=1)
+            out = self.diffusion_model(xc, t)
+        elif self.conditioning_key == 'crossattn':
+            cc = torch.cat(c_crossattn, 1)
+            out = self.diffusion_model(x, t, context=cc)
+        elif self.conditioning_key == 'hybrid':
+            xc = torch.cat([x] + c_concat, dim=1)
+            cc = torch.cat(c_crossattn, 1)
+            out = self.diffusion_model(xc, t, context=cc)
+        elif self.conditioning_key == 'adm':
+            cc = c_crossattn[0]
+            out = self.diffusion_model(x, t, y=cc)
+        else:
+            raise NotImplementedError()
+
+        return out
+
+
+class Layout2ImgDiffusionV1(LatentDiffusionV1):
+    # TODO: move all layout-specific hacks to this class
+    def __init__(self, cond_stage_key, *args, **kwargs):
+        assert cond_stage_key == 'coordinates_bbox', 'Layout2ImgDiffusion only for cond_stage_key="coordinates_bbox"'
+        super().__init__(*args, cond_stage_key=cond_stage_key, **kwargs)
+
+    def log_images(self, batch, N=8, *args, **kwargs):
+        logs = super().log_images(*args, batch=batch, N=N, **kwargs)
+
+        key = 'train' if self.training else 'validation'
+        dset = self.trainer.datamodule.datasets[key]
+        mapper = dset.conditional_builders[self.cond_stage_key]
+
+        bbox_imgs = []
+        map_fn = lambda catno: dset.get_textual_label(dset.get_category_id(catno))
+        for tknzd_bbox in batch[self.cond_stage_key][:N]:
+            bboximg = mapper.plot(tknzd_bbox.detach().cpu(), map_fn, (256, 256))
+            bbox_imgs.append(bboximg)
+
+        cond_img = torch.stack(bbox_imgs, dim=0)
+        logs['bbox_image'] = cond_img
+        return logs
+
+ldm.models.diffusion.ddpm.DDPMV1 = DDPMV1
+ldm.models.diffusion.ddpm.LatentDiffusionV1 = LatentDiffusionV1
+ldm.models.diffusion.ddpm.DiffusionWrapperV1 = DiffusionWrapperV1
+ldm.models.diffusion.ddpm.Layout2ImgDiffusionV1 = Layout2ImgDiffusionV1

extensions-builtin/LDSR/vqvae_quantize.py

@@ -0,0 +1,147 @@
+# Vendored from https://raw.githubusercontent.com/CompVis/taming-transformers/24268930bf1dce879235a7fddd0b2355b84d7ea6/taming/modules/vqvae/quantize.py,
+# where the license is as follows:
+#
+# Copyright (c) 2020 Patrick Esser and Robin Rombach and Björn Ommer
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+# EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+# IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
+# DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+# OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE
+# OR OTHER DEALINGS IN THE SOFTWARE./
+
+import torch
+import torch.nn as nn
+import numpy as np
+from einops import rearrange
+
+
+class VectorQuantizer2(nn.Module):
+    """
+    Improved version over VectorQuantizer, can be used as a drop-in replacement. Mostly
+    avoids costly matrix multiplications and allows for post-hoc remapping of indices.
+    """
+
+    # NOTE: due to a bug the beta term was applied to the wrong term. for
+    # backwards compatibility we use the buggy version by default, but you can
+    # specify legacy=False to fix it.
+    def __init__(self, n_e, e_dim, beta, remap=None, unknown_index="random",
+                 sane_index_shape=False, legacy=True):
+        super().__init__()
+        self.n_e = n_e
+        self.e_dim = e_dim
+        self.beta = beta
+        self.legacy = legacy
+
+        self.embedding = nn.Embedding(self.n_e, self.e_dim)
+        self.embedding.weight.data.uniform_(-1.0 / self.n_e, 1.0 / self.n_e)
+
+        self.remap = remap
+        if self.remap is not None:
+            self.register_buffer("used", torch.tensor(np.load(self.remap)))
+            self.re_embed = self.used.shape[0]
+            self.unknown_index = unknown_index  # "random" or "extra" or integer
+            if self.unknown_index == "extra":
+                self.unknown_index = self.re_embed
+                self.re_embed = self.re_embed + 1
+            print(f"Remapping {self.n_e} indices to {self.re_embed} indices. "
+                  f"Using {self.unknown_index} for unknown indices.")
+        else:
+            self.re_embed = n_e
+
+        self.sane_index_shape = sane_index_shape
+
+    def remap_to_used(self, inds):
+        ishape = inds.shape
+        assert len(ishape) > 1
+        inds = inds.reshape(ishape[0], -1)
+        used = self.used.to(inds)
+        match = (inds[:, :, None] == used[None, None, ...]).long()
+        new = match.argmax(-1)
+        unknown = match.sum(2) < 1
+        if self.unknown_index == "random":
+            new[unknown] = torch.randint(0, self.re_embed, size=new[unknown].shape).to(device=new.device)
+        else:
+            new[unknown] = self.unknown_index
+        return new.reshape(ishape)
+
+    def unmap_to_all(self, inds):
+        ishape = inds.shape
+        assert len(ishape) > 1
+        inds = inds.reshape(ishape[0], -1)
+        used = self.used.to(inds)
+        if self.re_embed > self.used.shape[0]:  # extra token
+            inds[inds >= self.used.shape[0]] = 0  # simply set to zero
+        back = torch.gather(used[None, :][inds.shape[0] * [0], :], 1, inds)
+        return back.reshape(ishape)
+
+    def forward(self, z, temp=None, rescale_logits=False, return_logits=False):
+        assert temp is None or temp == 1.0, "Only for interface compatible with Gumbel"
+        assert rescale_logits is False, "Only for interface compatible with Gumbel"
+        assert return_logits is False, "Only for interface compatible with Gumbel"
+        # reshape z -> (batch, height, width, channel) and flatten
+        z = rearrange(z, 'b c h w -> b h w c').contiguous()
+        z_flattened = z.view(-1, self.e_dim)
+        # distances from z to embeddings e_j (z - e)^2 = z^2 + e^2 - 2 e * z
+
+        d = torch.sum(z_flattened ** 2, dim=1, keepdim=True) + \
+            torch.sum(self.embedding.weight ** 2, dim=1) - 2 * \
+            torch.einsum('bd,dn->bn', z_flattened, rearrange(self.embedding.weight, 'n d -> d n'))
+
+        min_encoding_indices = torch.argmin(d, dim=1)
+        z_q = self.embedding(min_encoding_indices).view(z.shape)
+        perplexity = None
+        min_encodings = None
+
+        # compute loss for embedding
+        if not self.legacy:
+            loss = self.beta * torch.mean((z_q.detach() - z) ** 2) + \
+                   torch.mean((z_q - z.detach()) ** 2)
+        else:
+            loss = torch.mean((z_q.detach() - z) ** 2) + self.beta * \
+                   torch.mean((z_q - z.detach()) ** 2)
+
+        # preserve gradients
+        z_q = z + (z_q - z).detach()
+
+        # reshape back to match original input shape
+        z_q = rearrange(z_q, 'b h w c -> b c h w').contiguous()
+
+        if self.remap is not None:
+            min_encoding_indices = min_encoding_indices.reshape(z.shape[0], -1)  # add batch axis
+            min_encoding_indices = self.remap_to_used(min_encoding_indices)
+            min_encoding_indices = min_encoding_indices.reshape(-1, 1)  # flatten
+
+        if self.sane_index_shape:
+            min_encoding_indices = min_encoding_indices.reshape(
+                z_q.shape[0], z_q.shape[2], z_q.shape[3])
+
+        return z_q, loss, (perplexity, min_encodings, min_encoding_indices)
+
+    def get_codebook_entry(self, indices, shape):
+        # shape specifying (batch, height, width, channel)
+        if self.remap is not None:
+            indices = indices.reshape(shape[0], -1)  # add batch axis
+            indices = self.unmap_to_all(indices)
+            indices = indices.reshape(-1)  # flatten again
+
+        # get quantized latent vectors
+        z_q = self.embedding(indices)
+
+        if shape is not None:
+            z_q = z_q.view(shape)
+            # reshape back to match original input shape
+            z_q = z_q.permute(0, 3, 1, 2).contiguous()
+
+        return z_q

extensions-builtin/Lora/extra_networks_lora.py

@@ -0,0 +1,67 @@
+from modules import extra_networks, shared
+import networks
+
+
+class ExtraNetworkLora(extra_networks.ExtraNetwork):
+    def __init__(self):
+        super().__init__('lora')
+
+        self.errors = {}
+        """mapping of network names to the number of errors the network had during operation"""
+
+    def activate(self, p, params_list):
+        additional = shared.opts.sd_lora
+
+        self.errors.clear()
+
+        if additional != "None" and additional in networks.available_networks and not any(x for x in params_list if x.items[0] == additional):
+            p.all_prompts = [x + f"<lora:{additional}:{shared.opts.extra_networks_default_multiplier}>" for x in p.all_prompts]
+            params_list.append(extra_networks.ExtraNetworkParams(items=[additional, shared.opts.extra_networks_default_multiplier]))
+
+        names = []
+        te_multipliers = []
+        unet_multipliers = []
+        dyn_dims = []
+        for params in params_list:
+            assert params.items
+
+            names.append(params.positional[0])
+
+            te_multiplier = float(params.positional[1]) if len(params.positional) > 1 else 1.0
+            te_multiplier = float(params.named.get("te", te_multiplier))
+
+            unet_multiplier = float(params.positional[2]) if len(params.positional) > 2 else te_multiplier
+            unet_multiplier = float(params.named.get("unet", unet_multiplier))
+
+            dyn_dim = int(params.positional[3]) if len(params.positional) > 3 else None
+            dyn_dim = int(params.named["dyn"]) if "dyn" in params.named else dyn_dim
+
+            te_multipliers.append(te_multiplier)
+            unet_multipliers.append(unet_multiplier)
+            dyn_dims.append(dyn_dim)
+
+        networks.load_networks(names, te_multipliers, unet_multipliers, dyn_dims)
+
+        if shared.opts.lora_add_hashes_to_infotext:
+            network_hashes = []
+            for item in networks.loaded_networks:
+                shorthash = item.network_on_disk.shorthash
+                if not shorthash:
+                    continue
+
+                alias = item.mentioned_name
+                if not alias:
+                    continue
+
+                alias = alias.replace(":", "").replace(",", "")
+
+                network_hashes.append(f"{alias}: {shorthash}")
+
+            if network_hashes:
+                p.extra_generation_params["Lora hashes"] = ", ".join(network_hashes)
+
+    def deactivate(self, p):
+        if self.errors:
+            p.comment("Networks with errors: " + ", ".join(f"{k} ({v})" for k, v in self.errors.items()))
+
+            self.errors.clear()

extensions-builtin/Lora/lora.py

@@ -0,0 +1,9 @@
+import networks
+
+list_available_loras = networks.list_available_networks
+
+available_loras = networks.available_networks
+available_lora_aliases = networks.available_network_aliases
+available_lora_hash_lookup = networks.available_network_hash_lookup
+forbidden_lora_aliases = networks.forbidden_network_aliases
+loaded_loras = networks.loaded_networks

extensions-builtin/Lora/lora_logger.py

@@ -0,0 +1,33 @@
+import sys
+import copy
+import logging
+
+
+class ColoredFormatter(logging.Formatter):
+    COLORS = {
+        "DEBUG": "\033[0;36m",  # CYAN
+        "INFO": "\033[0;32m",  # GREEN
+        "WARNING": "\033[0;33m",  # YELLOW
+        "ERROR": "\033[0;31m",  # RED
+        "CRITICAL": "\033[0;37;41m",  # WHITE ON RED
+        "RESET": "\033[0m",  # RESET COLOR
+    }
+
+    def format(self, record):
+        colored_record = copy.copy(record)
+        levelname = colored_record.levelname
+        seq = self.COLORS.get(levelname, self.COLORS["RESET"])
+        colored_record.levelname = f"{seq}{levelname}{self.COLORS['RESET']}"
+        return super().format(colored_record)
+
+
+logger = logging.getLogger("lora")
+logger.propagate = False
+
+
+if not logger.handlers:
+    handler = logging.StreamHandler(sys.stdout)
+    handler.setFormatter(
+        ColoredFormatter("[%(name)s]-%(levelname)s: %(message)s")
+    )
+    logger.addHandler(handler)

extensions-builtin/Lora/lora_patches.py

@@ -0,0 +1,6 @@
+class LoraPatches:
+    def __init__(self):
+        pass
+
+    def undo(self):
+        pass

extensions-builtin/Lora/network.py

@@ -0,0 +1,190 @@
+from __future__ import annotations
+import os
+from collections import namedtuple
+import enum
+
+import torch.nn as nn
+import torch.nn.functional as F
+
+from modules import sd_models, cache, errors, hashes, shared
+
+NetworkWeights = namedtuple('NetworkWeights', ['network_key', 'sd_key', 'w', 'sd_module'])
+
+metadata_tags_order = {"ss_sd_model_name": 1, "ss_resolution": 2, "ss_clip_skip": 3, "ss_num_train_images": 10, "ss_tag_frequency": 20}
+
+
+class SdVersion(enum.Enum):
+    Unknown = 1
+    SD1 = 2
+    SD2 = 3
+    SDXL = 4
+
+
+class NetworkOnDisk:
+    def __init__(self, name, filename):
+        self.name = name
+        self.filename = filename
+        self.metadata = {}
+        self.is_safetensors = os.path.splitext(filename)[1].lower() == ".safetensors"
+
+        def read_metadata():
+            metadata = sd_models.read_metadata_from_safetensors(filename)
+            metadata.pop('ssmd_cover_images', None)  # those are cover images, and they are too big to display in UI as text
+
+            return metadata
+
+        if self.is_safetensors:
+            try:
+                self.metadata = cache.cached_data_for_file('safetensors-metadata', "lora/" + self.name, filename, read_metadata)
+            except Exception as e:
+                errors.display(e, f"reading lora {filename}")
+
+        if self.metadata:
+            m = {}
+            for k, v in sorted(self.metadata.items(), key=lambda x: metadata_tags_order.get(x[0], 999)):
+                m[k] = v
+
+            self.metadata = m
+
+        self.alias = self.metadata.get('ss_output_name', self.name)
+
+        self.hash = None
+        self.shorthash = None
+        self.set_hash(
+            self.metadata.get('sshs_model_hash') or
+            hashes.sha256_from_cache(self.filename, "lora/" + self.name, use_addnet_hash=self.is_safetensors) or
+            ''
+        )
+
+        self.sd_version = self.detect_version()
+
+    def detect_version(self):
+        if str(self.metadata.get('ss_base_model_version', "")).startswith("sdxl_"):
+            return SdVersion.SDXL
+        elif str(self.metadata.get('ss_v2', "")) == "True":
+            return SdVersion.SD2
+        elif len(self.metadata):
+            return SdVersion.SD1
+
+        return SdVersion.Unknown
+
+    def set_hash(self, v):
+        self.hash = v
+        self.shorthash = self.hash[0:12]
+
+        if self.shorthash:
+            import networks
+            networks.available_network_hash_lookup[self.shorthash] = self
+
+    def read_hash(self):
+        if not self.hash:
+            self.set_hash(hashes.sha256(self.filename, "lora/" + self.name, use_addnet_hash=self.is_safetensors) or '')
+
+    def get_alias(self):
+        import networks
+        if shared.opts.lora_preferred_name == "Filename" or self.alias.lower() in networks.forbidden_network_aliases:
+            return self.name
+        else:
+            return self.alias
+
+
+class Network:  # LoraModule
+    def __init__(self, name, network_on_disk: NetworkOnDisk):
+        self.name = name
+        self.network_on_disk = network_on_disk
+        self.te_multiplier = 1.0
+        self.unet_multiplier = 1.0
+        self.dyn_dim = None
+        self.modules = {}
+        self.bundle_embeddings = {}
+        self.mtime = None
+
+        self.mentioned_name = None
+        """the text that was used to add the network to prompt - can be either name or an alias"""
+
+
+class ModuleType:
+    def create_module(self, net: Network, weights: NetworkWeights) -> Network | None:
+        return None
+
+
+class NetworkModule:
+    def __init__(self, net: Network, weights: NetworkWeights):
+        self.network = net
+        self.network_key = weights.network_key
+        self.sd_key = weights.sd_key
+        self.sd_module = weights.sd_module
+
+        if hasattr(self.sd_module, 'weight'):
+            self.shape = self.sd_module.weight.shape
+
+        self.ops = None
+        self.extra_kwargs = {}
+        if isinstance(self.sd_module, nn.Conv2d):
+            self.ops = F.conv2d
+            self.extra_kwargs = {
+                'stride': self.sd_module.stride,
+                'padding': self.sd_module.padding
+            }
+        elif isinstance(self.sd_module, nn.Linear):
+            self.ops = F.linear
+        elif isinstance(self.sd_module, nn.LayerNorm):
+            self.ops = F.layer_norm
+            self.extra_kwargs = {
+                'normalized_shape': self.sd_module.normalized_shape,
+                'eps': self.sd_module.eps
+            }
+        elif isinstance(self.sd_module, nn.GroupNorm):
+            self.ops = F.group_norm
+            self.extra_kwargs = {
+                'num_groups': self.sd_module.num_groups,
+                'eps': self.sd_module.eps
+            }
+
+        self.dim = None
+        self.bias = weights.w.get("bias")
+        self.alpha = weights.w["alpha"].item() if "alpha" in weights.w else None
+        self.scale = weights.w["scale"].item() if "scale" in weights.w else None
+
+    def multiplier(self):
+        if 'transformer' in self.sd_key[:20]:
+            return self.network.te_multiplier
+        else:
+            return self.network.unet_multiplier
+
+    def calc_scale(self):
+        if self.scale is not None:
+            return self.scale
+        if self.dim is not None and self.alpha is not None:
+            return self.alpha / self.dim
+
+        return 1.0
+
+    def finalize_updown(self, updown, orig_weight, output_shape, ex_bias=None):
+        if self.bias is not None:
+            updown = updown.reshape(self.bias.shape)
+            updown += self.bias.to(orig_weight.device, dtype=updown.dtype)
+            updown = updown.reshape(output_shape)
+
+        if len(output_shape) == 4:
+            updown = updown.reshape(output_shape)
+
+        if orig_weight.size().numel() == updown.size().numel():
+            updown = updown.reshape(orig_weight.shape)
+
+        if ex_bias is not None:
+            ex_bias = ex_bias * self.multiplier()
+
+        return updown * self.calc_scale() * self.multiplier(), ex_bias
+
+    def calc_updown(self, target):
+        raise NotImplementedError()
+
+    def forward(self, x, y):
+        """A general forward implementation for all modules"""
+        if self.ops is None:
+            raise NotImplementedError()
+        else:
+            updown, ex_bias = self.calc_updown(self.sd_module.weight)
+            return y + self.ops(x, weight=updown, bias=ex_bias, **self.extra_kwargs)
+

extensions-builtin/Lora/networks.py

@@ -0,0 +1,219 @@
+import os
+import re
+
+import lora_patches
+import functools
+import network
+
+import torch
+from typing import Union
+
+from modules import shared, sd_models, errors, scripts
+from ldm_patched.modules.utils import load_torch_file
+from ldm_patched.modules.sd import load_lora_for_models
+
+
+@functools.lru_cache(maxsize=5)
+def load_lora_state_dict(filename):
+    return load_torch_file(filename, safe_load=True)
+
+
+def convert_diffusers_name_to_compvis(key, is_sd2):
+    pass
+
+
+def assign_network_names_to_compvis_modules(sd_model):
+    pass
+
+
+def load_network(name, network_on_disk):
+    net = network.Network(name, network_on_disk)
+    net.mtime = os.path.getmtime(network_on_disk.filename)
+
+    return net
+
+
+def purge_networks_from_memory():
+    pass
+
+
+def load_networks(names, te_multipliers=None, unet_multipliers=None, dyn_dims=None):
+    global lora_state_dict_cache
+
+    current_sd = sd_models.model_data.get_sd_model()
+    if current_sd is None:
+        return
+
+    loaded_networks.clear()
+
+    networks_on_disk = [available_networks.get(name, None) if name.lower() in forbidden_network_aliases else available_network_aliases.get(name, None) for name in names]
+    if any(x is None for x in networks_on_disk):
+        list_available_networks()
+        networks_on_disk = [available_networks.get(name, None) if name.lower() in forbidden_network_aliases else available_network_aliases.get(name, None) for name in names]
+
+    for i, (network_on_disk, name) in enumerate(zip(networks_on_disk, names)):
+        try:
+            net = load_network(name, network_on_disk)
+        except Exception as e:
+            errors.display(e, f"loading network {network_on_disk.filename}")
+            continue
+        net.mentioned_name = name
+        network_on_disk.read_hash()
+        loaded_networks.append(net)
+
+    compiled_lora_targets = []
+    for a, b, c in zip(networks_on_disk, unet_multipliers, te_multipliers):
+        compiled_lora_targets.append([a.filename, b, c])
+
+    compiled_lora_targets_hash = str(compiled_lora_targets)
+
+    if current_sd.current_lora_hash == compiled_lora_targets_hash:
+        return
+
+    current_sd.current_lora_hash = compiled_lora_targets_hash
+    current_sd.forge_objects.unet = current_sd.forge_objects_original.unet
+    current_sd.forge_objects.clip = current_sd.forge_objects_original.clip
+
+    for filename, strength_model, strength_clip in compiled_lora_targets:
+        lora_sd = load_lora_state_dict(filename)
+        current_sd.forge_objects.unet, current_sd.forge_objects.clip = load_lora_for_models(
+            current_sd.forge_objects.unet, current_sd.forge_objects.clip, lora_sd, strength_model, strength_clip,
+            filename=filename)
+
+    current_sd.forge_objects_after_applying_lora = current_sd.forge_objects.shallow_copy()
+    return
+
+
+def network_restore_weights_from_backup(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.GroupNorm, torch.nn.LayerNorm, torch.nn.MultiheadAttention]):
+    pass
+
+
+def network_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.GroupNorm, torch.nn.LayerNorm, torch.nn.MultiheadAttention]):
+    pass
+
+
+def network_forward(org_module, input, original_forward):
+    pass
+
+
+def network_reset_cached_weight(self: Union[torch.nn.Conv2d, torch.nn.Linear]):
+    pass
+
+
+def network_Linear_forward(self, input):
+    pass
+
+
+def network_Linear_load_state_dict(self, *args, **kwargs):
+    pass
+
+
+def network_Conv2d_forward(self, input):
+    pass
+
+
+def network_Conv2d_load_state_dict(self, *args, **kwargs):
+    pass
+
+
+def network_GroupNorm_forward(self, input):
+    pass
+
+
+def network_GroupNorm_load_state_dict(self, *args, **kwargs):
+    pass
+
+
+def network_LayerNorm_forward(self, input):
+    pass
+
+
+def network_LayerNorm_load_state_dict(self, *args, **kwargs):
+    pass
+
+
+def network_MultiheadAttention_forward(self, *args, **kwargs):
+    pass
+
+
+def network_MultiheadAttention_load_state_dict(self, *args, **kwargs):
+    pass
+
+
+def list_available_networks():
+    available_networks.clear()
+    available_network_aliases.clear()
+    forbidden_network_aliases.clear()
+    available_network_hash_lookup.clear()
+    forbidden_network_aliases.update({"none": 1, "Addams": 1})
+
+    os.makedirs(shared.cmd_opts.lora_dir, exist_ok=True)
+
+    candidates = list(shared.walk_files(shared.cmd_opts.lora_dir, allowed_extensions=[".pt", ".ckpt", ".safetensors"]))
+    for filename in candidates:
+        if os.path.isdir(filename):
+            continue
+
+        name = os.path.splitext(os.path.basename(filename))[0]
+        try:
+            entry = network.NetworkOnDisk(name, filename)
+        except OSError:  # should catch FileNotFoundError and PermissionError etc.
+            errors.report(f"Failed to load network {name} from {filename}", exc_info=True)
+            continue
+
+        available_networks[name] = entry
+
+        if entry.alias in available_network_aliases:
+            forbidden_network_aliases[entry.alias.lower()] = 1
+
+        available_network_aliases[name] = entry
+        available_network_aliases[entry.alias] = entry
+
+
+re_network_name = re.compile(r"(.*)\s*\([0-9a-fA-F]+\)")
+
+
+def infotext_pasted(infotext, params):
+    if "AddNet Module 1" in [x[1] for x in scripts.scripts_txt2img.infotext_fields]:
+        return  # if the other extension is active, it will handle those fields, no need to do anything
+
+    added = []
+
+    for k in params:
+        if not k.startswith("AddNet Model "):
+            continue
+
+        num = k[13:]
+
+        if params.get("AddNet Module " + num) != "LoRA":
+            continue
+
+        name = params.get("AddNet Model " + num)
+        if name is None:
+            continue
+
+        m = re_network_name.match(name)
+        if m:
+            name = m.group(1)
+
+        multiplier = params.get("AddNet Weight A " + num, "1.0")
+
+        added.append(f"<lora:{name}:{multiplier}>")
+
+    if added:
+        params["Prompt"] += "\n" + "".join(added)
+
+
+originals: lora_patches.LoraPatches = None
+
+extra_network_lora = None
+
+available_networks = {}
+available_network_aliases = {}
+loaded_networks = []
+loaded_bundle_embeddings = {}
+networks_in_memory = {}
+available_network_hash_lookup = {}
+forbidden_network_aliases = {}
+
+list_available_networks()

extensions-builtin/Lora/preload.py

@@ -0,0 +1,8 @@
+import os
+from modules import paths
+from modules.paths_internal import normalized_filepath
+
+
+def preload(parser):
+    parser.add_argument("--lora-dir", type=normalized_filepath, help="Path to directory with Lora networks.", default=os.path.join(paths.models_path, 'Lora'))
+    parser.add_argument("--lyco-dir-backcompat", type=normalized_filepath, help="Path to directory with LyCORIS networks (for backawards compatibility; can also use --lyco-dir).", default=os.path.join(paths.models_path, 'LyCORIS'))

extensions-builtin/Lora/scripts/lora_script.py

@@ -0,0 +1,100 @@
+import re
+
+import gradio as gr
+from fastapi import FastAPI
+
+import network
+import networks
+import lora  # noqa:F401
+import lora_patches
+import extra_networks_lora
+import ui_extra_networks_lora
+from modules import script_callbacks, ui_extra_networks, extra_networks, shared
+
+
+def unload():
+    networks.originals.undo()
+
+
+def before_ui():
+    ui_extra_networks.register_page(ui_extra_networks_lora.ExtraNetworksPageLora())
+
+    networks.extra_network_lora = extra_networks_lora.ExtraNetworkLora()
+    extra_networks.register_extra_network(networks.extra_network_lora)
+
+
+networks.originals = lora_patches.LoraPatches()
+
+script_callbacks.on_model_loaded(networks.assign_network_names_to_compvis_modules)
+script_callbacks.on_script_unloaded(unload)
+script_callbacks.on_before_ui(before_ui)
+script_callbacks.on_infotext_pasted(networks.infotext_pasted)
+
+
+shared.options_templates.update(shared.options_section(('extra_networks', "Extra Networks"), {
+    "sd_lora": shared.OptionInfo("None", "Add network to prompt", gr.Dropdown, lambda: {"choices": ["None", *networks.available_networks]}, refresh=networks.list_available_networks),
+    "lora_preferred_name": shared.OptionInfo("Alias from file", "When adding to prompt, refer to Lora by", gr.Radio, {"choices": ["Alias from file", "Filename"]}),
+    "lora_add_hashes_to_infotext": shared.OptionInfo(True, "Add Lora hashes to infotext"),
+    "lora_show_all": shared.OptionInfo(False, "Always show all networks on the Lora page").info("otherwise, those detected as for incompatible version of Stable Diffusion will be hidden"),
+    "lora_hide_unknown_for_versions": shared.OptionInfo([], "Hide networks of unknown versions for model versions", gr.CheckboxGroup, {"choices": ["SD1", "SD2", "SDXL"]}),
+    "lora_in_memory_limit": shared.OptionInfo(0, "Number of Lora networks to keep cached in memory", gr.Number, {"precision": 0}),
+    "lora_not_found_warning_console": shared.OptionInfo(False, "Lora not found warning in console"),
+    "lora_not_found_gradio_warning": shared.OptionInfo(False, "Lora not found warning popup in webui"),
+}))
+
+
+shared.options_templates.update(shared.options_section(('compatibility', "Compatibility"), {
+    "lora_functional": shared.OptionInfo(False, "Lora/Networks: use old method that takes longer when you have multiple Loras active and produces same results as kohya-ss/sd-webui-additional-networks extension"),
+}))
+
+
+def create_lora_json(obj: network.NetworkOnDisk):
+    return {
+        "name": obj.name,
+        "alias": obj.alias,
+        "path": obj.filename,
+        "metadata": obj.metadata,
+    }
+
+
+def api_networks(_: gr.Blocks, app: FastAPI):
+    @app.get("/sdapi/v1/loras")
+    async def get_loras():
+        return [create_lora_json(obj) for obj in networks.available_networks.values()]
+
+    @app.post("/sdapi/v1/refresh-loras")
+    async def refresh_loras():
+        return networks.list_available_networks()
+
+
+script_callbacks.on_app_started(api_networks)
+
+re_lora = re.compile("<lora:([^:]+):")
+
+
+def infotext_pasted(infotext, d):
+    hashes = d.get("Lora hashes")
+    if not hashes:
+        return
+
+    hashes = [x.strip().split(':', 1) for x in hashes.split(",")]
+    hashes = {x[0].strip().replace(",", ""): x[1].strip() for x in hashes}
+
+    def network_replacement(m):
+        alias = m.group(1)
+        shorthash = hashes.get(alias)
+        if shorthash is None:
+            return m.group(0)
+
+        network_on_disk = networks.available_network_hash_lookup.get(shorthash)
+        if network_on_disk is None:
+            return m.group(0)
+
+        return f'<lora:{network_on_disk.get_alias()}:'
+
+    d["Prompt"] = re.sub(re_lora, network_replacement, d["Prompt"])
+
+
+script_callbacks.on_infotext_pasted(infotext_pasted)
+
+shared.opts.onchange("lora_in_memory_limit", networks.purge_networks_from_memory)

extensions-builtin/Lora/ui_edit_user_metadata.py

@@ -0,0 +1,222 @@
+import datetime
+import html
+import random
+
+import gradio as gr
+import re
+
+from modules import ui_extra_networks_user_metadata
+
+
+def is_non_comma_tagset(tags):
+    average_tag_length = sum(len(x) for x in tags.keys()) / len(tags)
+
+    return average_tag_length >= 16
+
+
+re_word = re.compile(r"[-_\w']+")
+re_comma = re.compile(r" *, *")
+
+
+def build_tags(metadata):
+    tags = {}
+
+    for _, tags_dict in metadata.get("ss_tag_frequency", {}).items():
+        for tag, tag_count in tags_dict.items():
+            tag = tag.strip()
+            tags[tag] = tags.get(tag, 0) + int(tag_count)
+
+    if tags and is_non_comma_tagset(tags):
+        new_tags = {}
+
+        for text, text_count in tags.items():
+            for word in re.findall(re_word, text):
+                if len(word) < 3:
+                    continue
+
+                new_tags[word] = new_tags.get(word, 0) + text_count
+
+        tags = new_tags
+
+    ordered_tags = sorted(tags.keys(), key=tags.get, reverse=True)
+
+    return [(tag, tags[tag]) for tag in ordered_tags]
+
+
+class LoraUserMetadataEditor(ui_extra_networks_user_metadata.UserMetadataEditor):
+    def __init__(self, ui, tabname, page):
+        super().__init__(ui, tabname, page)
+
+        self.select_sd_version = None
+
+        self.taginfo = None
+        self.edit_activation_text = None
+        self.slider_preferred_weight = None
+        self.edit_notes = None
+
+    def save_lora_user_metadata(self, name, desc, sd_version, activation_text, preferred_weight, negative_text, notes):
+        user_metadata = self.get_user_metadata(name)
+        user_metadata["description"] = desc
+        user_metadata["sd version"] = sd_version
+        user_metadata["activation text"] = activation_text
+        user_metadata["preferred weight"] = preferred_weight
+        user_metadata["negative text"] = negative_text
+        user_metadata["notes"] = notes
+
+        self.write_user_metadata(name, user_metadata)
+
+    def get_metadata_table(self, name):
+        table = super().get_metadata_table(name)
+        item = self.page.items.get(name, {})
+        metadata = item.get("metadata") or {}
+
+        keys = {
+            'ss_output_name': "Output name:",
+            'ss_sd_model_name': "Model:",
+            'ss_clip_skip': "Clip skip:",
+            'ss_network_module': "Kohya module:",
+        }
+
+        for key, label in keys.items():
+            value = metadata.get(key, None)
+            if value is not None and str(value) != "None":
+                table.append((label, html.escape(value)))
+
+        ss_training_started_at = metadata.get('ss_training_started_at')
+        if ss_training_started_at:
+            table.append(("Date trained:", datetime.datetime.utcfromtimestamp(float(ss_training_started_at)).strftime('%Y-%m-%d %H:%M')))
+
+        ss_bucket_info = metadata.get("ss_bucket_info")
+        if ss_bucket_info and "buckets" in ss_bucket_info:
+            resolutions = {}
+            for _, bucket in ss_bucket_info["buckets"].items():
+                resolution = bucket["resolution"]
+                resolution = f'{resolution[1]}x{resolution[0]}'
+
+                resolutions[resolution] = resolutions.get(resolution, 0) + int(bucket["count"])
+
+            resolutions_list = sorted(resolutions.keys(), key=resolutions.get, reverse=True)
+            resolutions_text = html.escape(", ".join(resolutions_list[0:4]))
+            if len(resolutions) > 4:
+                resolutions_text += ", ..."
+                resolutions_text = f"<span title='{html.escape(', '.join(resolutions_list))}'>{resolutions_text}</span>"
+
+            table.append(('Resolutions:' if len(resolutions_list) > 1 else 'Resolution:', resolutions_text))
+
+        image_count = 0
+        for _, params in metadata.get("ss_dataset_dirs", {}).items():
+            image_count += int(params.get("img_count", 0))
+
+        if image_count:
+            table.append(("Dataset size:", image_count))
+
+        return table
+
+    def put_values_into_components(self, name):
+        user_metadata = self.get_user_metadata(name)
+        values = super().put_values_into_components(name)
+
+        item = self.page.items.get(name, {})
+        metadata = item.get("metadata") or {}
+
+        tags = build_tags(metadata)
+        gradio_tags = [(tag, str(count)) for tag, count in tags[0:24]]
+
+        return [
+            *values[0:5],
+            item.get("sd_version", "Unknown"),
+            gr.HighlightedText.update(value=gradio_tags, visible=True if tags else False),
+            user_metadata.get('activation text', ''),
+            float(user_metadata.get('preferred weight', 0.0)),
+            user_metadata.get('negative text', ''),
+            gr.update(visible=True if tags else False),
+            gr.update(value=self.generate_random_prompt_from_tags(tags), visible=True if tags else False),
+        ]
+
+    def generate_random_prompt(self, name):
+        item = self.page.items.get(name, {})
+        metadata = item.get("metadata") or {}
+        tags = build_tags(metadata)
+
+        return self.generate_random_prompt_from_tags(tags)
+
+    def generate_random_prompt_from_tags(self, tags):
+        max_count = None
+        res = []
+        for tag, count in tags:
+            if not max_count:
+                max_count = count
+
+            v = random.random() * max_count
+            if count > v:
+                res.append(tag)
+
+        return ", ".join(sorted(res))
+
+    def create_extra_default_items_in_left_column(self):
+
+        # this would be a lot better as gr.Radio but I can't make it work
+        self.select_sd_version = gr.Dropdown(['SD1', 'SD2', 'SDXL', 'Unknown'], value='Unknown', label='Stable Diffusion version', interactive=True)
+
+    def create_editor(self):
+        self.create_default_editor_elems()
+
+        self.taginfo = gr.HighlightedText(label="Training dataset tags")
+        self.edit_activation_text = gr.Text(label='Activation text', info="Will be added to prompt along with Lora")
+        self.slider_preferred_weight = gr.Slider(label='Preferred weight', info="Set to 0 to disable", minimum=0.0, maximum=2.0, step=0.01)
+        self.edit_negative_text = gr.Text(label='Negative prompt', info="Will be added to negative prompts")
+        with gr.Row() as row_random_prompt:
+            with gr.Column(scale=8):
+                random_prompt = gr.Textbox(label='Random prompt', lines=4, max_lines=4, interactive=False)
+
+            with gr.Column(scale=1, min_width=120):
+                generate_random_prompt = gr.Button('Generate', size="lg", scale=1)
+
+        self.edit_notes = gr.TextArea(label='Notes', lines=4)
+
+        generate_random_prompt.click(fn=self.generate_random_prompt, inputs=[self.edit_name_input], outputs=[random_prompt], show_progress=False)
+
+        def select_tag(activation_text, evt: gr.SelectData):
+            tag = evt.value[0]
+
+            words = re.split(re_comma, activation_text)
+            if tag in words:
+                words = [x for x in words if x != tag and x.strip()]
+                return ", ".join(words)
+
+            return activation_text + ", " + tag if activation_text else tag
+
+        self.taginfo.select(fn=select_tag, inputs=[self.edit_activation_text], outputs=[self.edit_activation_text], show_progress=False)
+
+        self.create_default_buttons()
+
+        viewed_components = [
+            self.edit_name,
+            self.edit_description,
+            self.html_filedata,
+            self.html_preview,
+            self.edit_notes,
+            self.select_sd_version,
+            self.taginfo,
+            self.edit_activation_text,
+            self.slider_preferred_weight,
+            self.edit_negative_text,
+            row_random_prompt,
+            random_prompt,
+        ]
+
+        self.button_edit\
+            .click(fn=self.put_values_into_components, inputs=[self.edit_name_input], outputs=viewed_components)\
+            .then(fn=lambda: gr.update(visible=True), inputs=[], outputs=[self.box])
+
+        edited_components = [
+            self.edit_description,
+            self.select_sd_version,
+            self.edit_activation_text,
+            self.slider_preferred_weight,
+            self.edit_negative_text,
+            self.edit_notes,
+        ]
+
+
+        self.setup_save_handler(self.button_save, self.save_lora_user_metadata, edited_components)

extensions-builtin/Lora/ui_extra_networks_lora.py

@@ -0,0 +1,90 @@
+import os
+
+import network
+import networks
+
+from modules import shared, ui_extra_networks
+from modules.ui_extra_networks import quote_js
+from ui_edit_user_metadata import LoraUserMetadataEditor
+
+
+class ExtraNetworksPageLora(ui_extra_networks.ExtraNetworksPage):
+    def __init__(self):
+        super().__init__('Lora')
+
+    def refresh(self):
+        networks.list_available_networks()
+
+    def create_item(self, name, index=None, enable_filter=True):
+        lora_on_disk = networks.available_networks.get(name)
+        if lora_on_disk is None:
+            return
+
+        path, ext = os.path.splitext(lora_on_disk.filename)
+
+        alias = lora_on_disk.get_alias()
+
+        search_terms = [self.search_terms_from_path(lora_on_disk.filename)]
+        if lora_on_disk.hash:
+            search_terms.append(lora_on_disk.hash)
+        item = {
+            "name": name,
+            "filename": lora_on_disk.filename,
+            "shorthash": lora_on_disk.shorthash,
+            "preview": self.find_preview(path),
+            "description": self.find_description(path),
+            "search_terms": search_terms,
+            "local_preview": f"{path}.{shared.opts.samples_format}",
+            "metadata": lora_on_disk.metadata,
+            "sort_keys": {'default': index, **self.get_sort_keys(lora_on_disk.filename)},
+            "sd_version": lora_on_disk.sd_version.name,
+        }
+
+        self.read_user_metadata(item)
+        activation_text = item["user_metadata"].get("activation text")
+        preferred_weight = item["user_metadata"].get("preferred weight", 0.0)
+        item["prompt"] = quote_js(f"<lora:{alias}:") + " + " + (str(preferred_weight) if preferred_weight else "opts.extra_networks_default_multiplier") + " + " + quote_js(">")
+
+        if activation_text:
+            item["prompt"] += " + " + quote_js(" " + activation_text)
+
+        negative_prompt = item["user_metadata"].get("negative text")
+        item["negative_prompt"] = quote_js("")
+        if negative_prompt:
+            item["negative_prompt"] = quote_js('(' + negative_prompt + ':1)')
+
+        sd_version = item["user_metadata"].get("sd version")
+        if sd_version in network.SdVersion.__members__:
+            item["sd_version"] = sd_version
+            sd_version = network.SdVersion[sd_version]
+        else:
+            sd_version = lora_on_disk.sd_version
+
+        if shared.opts.lora_show_all or not enable_filter:
+            pass
+        elif sd_version == network.SdVersion.Unknown:
+            model_version = network.SdVersion.SDXL if shared.sd_model.is_sdxl else network.SdVersion.SD2 if shared.sd_model.is_sd2 else network.SdVersion.SD1
+            if model_version.name in shared.opts.lora_hide_unknown_for_versions:
+                return None
+        elif shared.sd_model.is_sdxl and sd_version != network.SdVersion.SDXL:
+            return None
+        elif shared.sd_model.is_sd2 and sd_version != network.SdVersion.SD2:
+            return None
+        elif shared.sd_model.is_sd1 and sd_version != network.SdVersion.SD1:
+            return None
+
+        return item
+
+    def list_items(self):
+        # instantiate a list to protect against concurrent modification
+        names = list(networks.available_networks)
+        for index, name in enumerate(names):
+            item = self.create_item(name, index)
+            if item is not None:
+                yield item
+
+    def allowed_directories_for_previews(self):
+        return [shared.cmd_opts.lora_dir]
+
+    def create_user_metadata_editor(self, ui, tabname):
+        return LoraUserMetadataEditor(ui, tabname, self)

extensions-builtin/ScuNET/preload.py

@@ -0,0 +1,6 @@
+import os
+from modules import paths
+
+
+def preload(parser):
+    parser.add_argument("--scunet-models-path", type=str, help="Path to directory with ScuNET model file(s).", default=os.path.join(paths.models_path, 'ScuNET'))

extensions-builtin/ScuNET/scripts/scunet_model.py

@@ -0,0 +1,74 @@
+import sys
+
+import PIL.Image
+
+import modules.upscaler
+from modules import devices, errors, modelloader, script_callbacks, shared, upscaler_utils
+
+
+class UpscalerScuNET(modules.upscaler.Upscaler):
+    def __init__(self, dirname):
+        self.name = "ScuNET"
+        self.model_name = "ScuNET GAN"
+        self.model_name2 = "ScuNET PSNR"
+        self.model_url = "https://github.com/cszn/KAIR/releases/download/v1.0/scunet_color_real_gan.pth"
+        self.model_url2 = "https://github.com/cszn/KAIR/releases/download/v1.0/scunet_color_real_psnr.pth"
+        self.user_path = dirname
+        super().__init__()
+        model_paths = self.find_models(ext_filter=[".pth"])
+        scalers = []
+        add_model2 = True
+        for file in model_paths:
+            if file.startswith("http"):
+                name = self.model_name
+            else:
+                name = modelloader.friendly_name(file)
+            if name == self.model_name2 or file == self.model_url2:
+                add_model2 = False
+            try:
+                scaler_data = modules.upscaler.UpscalerData(name, file, self, 4)
+                scalers.append(scaler_data)
+            except Exception:
+                errors.report(f"Error loading ScuNET model: {file}", exc_info=True)
+        if add_model2:
+            scaler_data2 = modules.upscaler.UpscalerData(self.model_name2, self.model_url2, self)
+            scalers.append(scaler_data2)
+        self.scalers = scalers
+
+    def do_upscale(self, img: PIL.Image.Image, selected_file):
+        devices.torch_gc()
+        try:
+            model = self.load_model(selected_file)
+        except Exception as e:
+            print(f"ScuNET: Unable to load model from {selected_file}: {e}", file=sys.stderr)
+            return img
+
+        img = upscaler_utils.upscale_2(
+            img,
+            model,
+            tile_size=shared.opts.SCUNET_tile,
+            tile_overlap=shared.opts.SCUNET_tile_overlap,
+            scale=1,  # ScuNET is a denoising model, not an upscaler
+            desc='ScuNET',
+        )
+        devices.torch_gc()
+        return img
+
+    def load_model(self, path: str):
+        device = devices.get_device_for('scunet')
+        if path.startswith("http"):
+            # TODO: this doesn't use `path` at all?
+            filename = modelloader.load_file_from_url(self.model_url, model_dir=self.model_download_path, file_name=f"{self.name}.pth")
+        else:
+            filename = path
+        return modelloader.load_spandrel_model(filename, device=device, expected_architecture='SCUNet')
+
+
+def on_ui_settings():
+    import gradio as gr
+
+    shared.opts.add_option("SCUNET_tile", shared.OptionInfo(256, "Tile size for SCUNET upscalers.", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}, section=('upscaling', "Upscaling")).info("0 = no tiling"))
+    shared.opts.add_option("SCUNET_tile_overlap", shared.OptionInfo(8, "Tile overlap for SCUNET upscalers.", gr.Slider, {"minimum": 0, "maximum": 64, "step": 1}, section=('upscaling', "Upscaling")).info("Low values = visible seam"))
+
+
+script_callbacks.on_ui_settings(on_ui_settings)

extensions-builtin/SwinIR/preload.py

@@ -0,0 +1,6 @@
+import os
+from modules import paths
+
+
+def preload(parser):
+    parser.add_argument("--swinir-models-path", type=str, help="Path to directory with SwinIR model file(s).", default=os.path.join(paths.models_path, 'SwinIR'))

extensions-builtin/SwinIR/scripts/swinir_model.py

@@ -0,0 +1,98 @@
+import logging
+import sys
+
+import torch
+from PIL import Image
+
+from modules import devices, modelloader, script_callbacks, shared, upscaler_utils
+from modules.upscaler import Upscaler, UpscalerData
+from modules_forge.forge_util import prepare_free_memory
+
+SWINIR_MODEL_URL = "https://github.com/JingyunLiang/SwinIR/releases/download/v0.0/003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR-L_x4_GAN.pth"
+
+logger = logging.getLogger(__name__)
+
+
+class UpscalerSwinIR(Upscaler):
+    def __init__(self, dirname):
+        self._cached_model = None           # keep the model when SWIN_torch_compile is on to prevent re-compile every runs
+        self._cached_model_config = None    # to clear '_cached_model' when changing model (v1/v2) or settings
+        self.name = "SwinIR"
+        self.model_url = SWINIR_MODEL_URL
+        self.model_name = "SwinIR 4x"
+        self.user_path = dirname
+        super().__init__()
+        scalers = []
+        model_files = self.find_models(ext_filter=[".pt", ".pth"])
+        for model in model_files:
+            if model.startswith("http"):
+                name = self.model_name
+            else:
+                name = modelloader.friendly_name(model)
+            model_data = UpscalerData(name, model, self)
+            scalers.append(model_data)
+        self.scalers = scalers
+
+    def do_upscale(self, img: Image.Image, model_file: str) -> Image.Image:
+        prepare_free_memory()
+
+        current_config = (model_file, shared.opts.SWIN_tile)
+
+        if self._cached_model_config == current_config:
+            model = self._cached_model
+        else:
+            try:
+                model = self.load_model(model_file)
+            except Exception as e:
+                print(f"Failed loading SwinIR model {model_file}: {e}", file=sys.stderr)
+                return img
+            self._cached_model = model
+            self._cached_model_config = current_config
+
+        img = upscaler_utils.upscale_2(
+            img,
+            model,
+            tile_size=shared.opts.SWIN_tile,
+            tile_overlap=shared.opts.SWIN_tile_overlap,
+            scale=model.scale,
+            desc="SwinIR",
+        )
+        devices.torch_gc()
+        return img
+
+    def load_model(self, path, scale=4):
+        if path.startswith("http"):
+            filename = modelloader.load_file_from_url(
+                url=path,
+                model_dir=self.model_download_path,
+                file_name=f"{self.model_name.replace(' ', '_')}.pth",
+            )
+        else:
+            filename = path
+
+        model_descriptor = modelloader.load_spandrel_model(
+            filename,
+            device=self._get_device(),
+            prefer_half=(devices.dtype == torch.float16),
+            expected_architecture="SwinIR",
+        )
+        if getattr(shared.opts, 'SWIN_torch_compile', False):
+            try:
+                model_descriptor.model.compile()
+            except Exception:
+                logger.warning("Failed to compile SwinIR model, fallback to JIT", exc_info=True)
+        return model_descriptor
+
+    def _get_device(self):
+        return devices.get_device_for('swinir')
+
+
+def on_ui_settings():
+    import gradio as gr
+
+    shared.opts.add_option("SWIN_tile", shared.OptionInfo(192, "Tile size for all SwinIR.", gr.Slider, {"minimum": 16, "maximum": 512, "step": 16}, section=('upscaling', "Upscaling")))
+    shared.opts.add_option("SWIN_tile_overlap", shared.OptionInfo(8, "Tile overlap, in pixels for SwinIR. Low values = visible seam.", gr.Slider, {"minimum": 0, "maximum": 48, "step": 1}, section=('upscaling', "Upscaling")))
+    shared.opts.add_option("SWIN_torch_compile", shared.OptionInfo(False, "Use torch.compile to accelerate SwinIR.", gr.Checkbox, {"interactive": True}, section=('upscaling', "Upscaling")).info("Takes longer on first run"))
+
+
+script_callbacks.on_ui_settings(on_ui_settings)

extensions-builtin/canvas-zoom-and-pan/javascript/zoom.js

@@ -0,0 +1,968 @@
+onUiLoaded(async() => {
+    const elementIDs = {
+        img2imgTabs: "#mode_img2img .tab-nav",
+        inpaint: "#img2maskimg",
+        inpaintSketch: "#inpaint_sketch",
+        rangeGroup: "#img2img_column_size",
+        sketch: "#img2img_sketch"
+    };
+    const tabNameToElementId = {
+        "Inpaint sketch": elementIDs.inpaintSketch,
+        "Inpaint": elementIDs.inpaint,
+        "Sketch": elementIDs.sketch
+    };
+
+
+    // Helper functions
+    // Get active tab
+
+    /**
+     * Waits for an element to be present in the DOM.
+     */
+    const waitForElement = (id) => new Promise(resolve => {
+        const checkForElement = () => {
+            const element = document.querySelector(id);
+            if (element) return resolve(element);
+            setTimeout(checkForElement, 100);
+        };
+        checkForElement();
+    });
+
+    function getActiveTab(elements, all = false) {
+        const tabs = elements.img2imgTabs.querySelectorAll("button");
+
+        if (all) return tabs;
+
+        for (let tab of tabs) {
+            if (tab.classList.contains("selected")) {
+                return tab;
+            }
+        }
+    }
+
+    // Get tab ID
+    function getTabId(elements) {
+        const activeTab = getActiveTab(elements);
+        return tabNameToElementId[activeTab.innerText];
+    }
+
+    // Wait until opts loaded
+    async function waitForOpts() {
+        for (; ;) {
+            if (window.opts && Object.keys(window.opts).length) {
+                return window.opts;
+            }
+            await new Promise(resolve => setTimeout(resolve, 100));
+        }
+    }
+
+    // Detect whether the element has a horizontal scroll bar
+    function hasHorizontalScrollbar(element) {
+        return element.scrollWidth > element.clientWidth;
+    }
+
+    // Function for defining the "Ctrl", "Shift" and "Alt" keys
+    function isModifierKey(event, key) {
+        switch (key) {
+        case "Ctrl":
+            return event.ctrlKey;
+        case "Shift":
+            return event.shiftKey;
+        case "Alt":
+            return event.altKey;
+        default:
+            return false;
+        }
+    }
+
+    // Check if hotkey is valid
+    function isValidHotkey(value) {
+        const specialKeys = ["Ctrl", "Alt", "Shift", "Disable"];
+        return (
+            (typeof value === "string" &&
+                value.length === 1 &&
+                /[a-z]/i.test(value)) ||
+            specialKeys.includes(value)
+        );
+    }
+
+    // Normalize hotkey
+    function normalizeHotkey(hotkey) {
+        return hotkey.length === 1 ? "Key" + hotkey.toUpperCase() : hotkey;
+    }
+
+    // Format hotkey for display
+    function formatHotkeyForDisplay(hotkey) {
+        return hotkey.startsWith("Key") ? hotkey.slice(3) : hotkey;
+    }
+
+    // Create hotkey configuration with the provided options
+    function createHotkeyConfig(defaultHotkeysConfig, hotkeysConfigOpts) {
+        const result = {}; // Resulting hotkey configuration
+        const usedKeys = new Set(); // Set of used hotkeys
+
+        // Iterate through defaultHotkeysConfig keys
+        for (const key in defaultHotkeysConfig) {
+            const userValue = hotkeysConfigOpts[key]; // User-provided hotkey value
+            const defaultValue = defaultHotkeysConfig[key]; // Default hotkey value
+
+            // Apply appropriate value for undefined, boolean, or object userValue
+            if (
+                userValue === undefined ||
+                typeof userValue === "boolean" ||
+                typeof userValue === "object" ||
+                userValue === "disable"
+            ) {
+                result[key] =
+                    userValue === undefined ? defaultValue : userValue;
+            } else if (isValidHotkey(userValue)) {
+                const normalizedUserValue = normalizeHotkey(userValue);
+
+                // Check for conflicting hotkeys
+                if (!usedKeys.has(normalizedUserValue)) {
+                    usedKeys.add(normalizedUserValue);
+                    result[key] = normalizedUserValue;
+                } else {
+                    console.error(
+                        `Hotkey: ${formatHotkeyForDisplay(
+                            userValue
+                        )} for ${key} is repeated and conflicts with another hotkey. The default hotkey is used: ${formatHotkeyForDisplay(
+                            defaultValue
+                        )}`
+                    );
+                    result[key] = defaultValue;
+                }
+            } else {
+                console.error(
+                    `Hotkey: ${formatHotkeyForDisplay(
+                        userValue
+                    )} for ${key} is not valid. The default hotkey is used: ${formatHotkeyForDisplay(
+                        defaultValue
+                    )}`
+                );
+                result[key] = defaultValue;
+            }
+        }
+
+        return result;
+    }
+
+    // Disables functions in the config object based on the provided list of function names
+    function disableFunctions(config, disabledFunctions) {
+        // Bind the hasOwnProperty method to the functionMap object to avoid errors
+        const hasOwnProperty =
+            Object.prototype.hasOwnProperty.bind(functionMap);
+
+        // Loop through the disabledFunctions array and disable the corresponding functions in the config object
+        disabledFunctions.forEach(funcName => {
+            if (hasOwnProperty(funcName)) {
+                const key = functionMap[funcName];
+                config[key] = "disable";
+            }
+        });
+
+        // Return the updated config object
+        return config;
+    }
+
+    /**
+     * The restoreImgRedMask function displays a red mask around an image to indicate the aspect ratio.
+     * If the image display property is set to 'none', the mask breaks. To fix this, the function
+     * temporarily sets the display property to 'block' and then hides the mask again after 300 milliseconds
+     * to avoid breaking the canvas. Additionally, the function adjusts the mask to work correctly on
+     * very long images.
+     */
+    function restoreImgRedMask(elements) {
+        const mainTabId = getTabId(elements);
+
+        if (!mainTabId) return;
+
+        const mainTab = gradioApp().querySelector(mainTabId);
+        const img = mainTab.querySelector("img");
+        const imageARPreview = gradioApp().querySelector("#imageARPreview");
+
+        if (!img || !imageARPreview) return;
+
+        imageARPreview.style.transform = "";
+        if (parseFloat(mainTab.style.width) > 865) {
+            const transformString = mainTab.style.transform;
+            const scaleMatch = transformString.match(
+                /scale\(([-+]?[0-9]*\.?[0-9]+)\)/
+            );
+            let zoom = 1; // default zoom
+
+            if (scaleMatch && scaleMatch[1]) {
+                zoom = Number(scaleMatch[1]);
+            }
+
+            imageARPreview.style.transformOrigin = "0 0";
+            imageARPreview.style.transform = `scale(${zoom})`;
+        }
+
+        if (img.style.display !== "none") return;
+
+        img.style.display = "block";
+
+        setTimeout(() => {
+            img.style.display = "none";
+        }, 400);
+    }
+
+    const hotkeysConfigOpts = await waitForOpts();
+
+    // Default config
+    const defaultHotkeysConfig = {
+        canvas_hotkey_zoom: "Alt",
+        canvas_hotkey_adjust: "Ctrl",
+        canvas_hotkey_reset: "KeyR",
+        canvas_hotkey_fullscreen: "KeyS",
+        canvas_hotkey_move: "KeyF",
+        canvas_hotkey_overlap: "KeyO",
+        canvas_hotkey_shrink_brush: "KeyQ",
+        canvas_hotkey_grow_brush: "KeyW",
+        canvas_disabled_functions: [],
+        canvas_show_tooltip: true,
+        canvas_auto_expand: true,
+        canvas_blur_prompt: false,
+    };
+
+    const functionMap = {
+        "Zoom": "canvas_hotkey_zoom",
+        "Adjust brush size": "canvas_hotkey_adjust",
+        "Hotkey shrink brush": "canvas_hotkey_shrink_brush",
+        "Hotkey enlarge brush": "canvas_hotkey_grow_brush",
+        "Moving canvas": "canvas_hotkey_move",
+        "Fullscreen": "canvas_hotkey_fullscreen",
+        "Reset Zoom": "canvas_hotkey_reset",
+        "Overlap": "canvas_hotkey_overlap"
+    };
+
+    // Loading the configuration from opts
+    const preHotkeysConfig = createHotkeyConfig(
+        defaultHotkeysConfig,
+        hotkeysConfigOpts
+    );
+
+    // Disable functions that are not needed by the user
+    const hotkeysConfig = disableFunctions(
+        preHotkeysConfig,
+        preHotkeysConfig.canvas_disabled_functions
+    );
+
+    let isMoving = false;
+    let mouseX, mouseY;
+    let activeElement;
+
+    const elements = Object.fromEntries(
+        Object.keys(elementIDs).map(id => [
+            id,
+            gradioApp().querySelector(elementIDs[id])
+        ])
+    );
+    const elemData = {};
+
+    // Apply functionality to the range inputs. Restore redmask and correct for long images.
+    const rangeInputs = elements.rangeGroup ?
+        Array.from(elements.rangeGroup.querySelectorAll("input")) :
+        [
+            gradioApp().querySelector("#img2img_width input[type='range']"),
+            gradioApp().querySelector("#img2img_height input[type='range']")
+        ];
+
+    for (const input of rangeInputs) {
+        input?.addEventListener("input", () => restoreImgRedMask(elements));
+    }
+
+    function applyZoomAndPan(elemId, isExtension = true) {
+        const targetElement = gradioApp().querySelector(elemId);
+
+        if (!targetElement) {
+            console.log("Element not found");
+            return;
+        }
+
+        targetElement.style.transformOrigin = "0 0";
+
+        elemData[elemId] = {
+            zoom: 1,
+            panX: 0,
+            panY: 0
+        };
+        let fullScreenMode = false;
+
+        // Create tooltip
+        function createTooltip() {
+            const toolTipElemnt =
+                targetElement.querySelector(".image-container");
+            const tooltip = document.createElement("div");
+            tooltip.className = "canvas-tooltip";
+
+            // Creating an item of information
+            const info = document.createElement("i");
+            info.className = "canvas-tooltip-info";
+            info.textContent = "";
+
+            // Create a container for the contents of the tooltip
+            const tooltipContent = document.createElement("div");
+            tooltipContent.className = "canvas-tooltip-content";
+
+            // Define an array with hotkey information and their actions
+            const hotkeysInfo = [
+                {
+                    configKey: "canvas_hotkey_zoom",
+                    action: "Zoom canvas",
+                    keySuffix: " + wheel"
+                },
+                {
+                    configKey: "canvas_hotkey_adjust",
+                    action: "Adjust brush size",
+                    keySuffix: " + wheel"
+                },
+                {configKey: "canvas_hotkey_reset", action: "Reset zoom"},
+                {
+                    configKey: "canvas_hotkey_fullscreen",
+                    action: "Fullscreen mode"
+                },
+                {configKey: "canvas_hotkey_move", action: "Move canvas"},
+                {configKey: "canvas_hotkey_overlap", action: "Overlap"}
+            ];
+
+            // Create hotkeys array with disabled property based on the config values
+            const hotkeys = hotkeysInfo.map(info => {
+                const configValue = hotkeysConfig[info.configKey];
+                const key = info.keySuffix ?
+                    `${configValue}${info.keySuffix}` :
+                    configValue.charAt(configValue.length - 1);
+                return {
+                    key,
+                    action: info.action,
+                    disabled: configValue === "disable"
+                };
+            });
+
+            for (const hotkey of hotkeys) {
+                if (hotkey.disabled) {
+                    continue;
+                }
+
+                const p = document.createElement("p");
+                p.innerHTML = `<b>${hotkey.key}</b> - ${hotkey.action}`;
+                tooltipContent.appendChild(p);
+            }
+
+            // Add information and content elements to the tooltip element
+            tooltip.appendChild(info);
+            tooltip.appendChild(tooltipContent);
+
+            // Add a hint element to the target element
+            toolTipElemnt.appendChild(tooltip);
+        }
+
+        //Show tool tip if setting enable
+        if (hotkeysConfig.canvas_show_tooltip) {
+            createTooltip();
+        }
+
+        // In the course of research, it was found that the tag img is very harmful when zooming and creates white canvases. This hack allows you to almost never think about this problem, it has no effect on webui.
+        function fixCanvas() {
+            const activeTab = getActiveTab(elements).textContent.trim();
+
+            if (activeTab !== "img2img") {
+                const img = targetElement.querySelector(`${elemId} img`);
+
+                if (img && img.style.display !== "none") {
+                    img.style.display = "none";
+                    img.style.visibility = "hidden";
+                }
+            }
+        }
+
+        // Reset the zoom level and pan position of the target element to their initial values
+        function resetZoom() {
+            elemData[elemId] = {
+                zoomLevel: 1,
+                panX: 0,
+                panY: 0
+            };
+
+            if (isExtension) {
+                targetElement.style.overflow = "hidden";
+            }
+
+            targetElement.isZoomed = false;
+
+            fixCanvas();
+            targetElement.style.transform = `scale(${elemData[elemId].zoomLevel}) translate(${elemData[elemId].panX}px, ${elemData[elemId].panY}px)`;
+
+            const canvas = gradioApp().querySelector(
+                `${elemId} canvas[key="interface"]`
+            );
+
+            toggleOverlap("off");
+            fullScreenMode = false;
+
+            const closeBtn = targetElement.querySelector("button[aria-label='Remove Image']");
+            if (closeBtn) {
+                closeBtn.addEventListener("click", resetZoom);
+            }
+
+            if (canvas && isExtension) {
+                const parentElement = targetElement.closest('[id^="component-"]');
+                if (
+                    canvas &&
+                    parseFloat(canvas.style.width) > parentElement.offsetWidth &&
+                    parseFloat(targetElement.style.width) > parentElement.offsetWidth
+                ) {
+                    fitToElement();
+                    return;
+                }
+
+            }
+
+            if (
+                canvas &&
+                !isExtension &&
+                parseFloat(canvas.style.width) > 865 &&
+                parseFloat(targetElement.style.width) > 865
+            ) {
+                fitToElement();
+                return;
+            }
+
+            targetElement.style.width = "";
+        }
+
+        // Toggle the zIndex of the target element between two values, allowing it to overlap or be overlapped by other elements
+        function toggleOverlap(forced = "") {
+            const zIndex1 = "0";
+            const zIndex2 = "998";
+
+            targetElement.style.zIndex =
+                targetElement.style.zIndex !== zIndex2 ? zIndex2 : zIndex1;
+
+            if (forced === "off") {
+                targetElement.style.zIndex = zIndex1;
+            } else if (forced === "on") {
+                targetElement.style.zIndex = zIndex2;
+            }
+        }
+
+        // Adjust the brush size based on the deltaY value from a mouse wheel event
+        function adjustBrushSize(
+            elemId,
+            deltaY,
+            withoutValue = false,
+            percentage = 5
+        ) {
+            const input =
+                gradioApp().querySelector(
+                    `${elemId} input[aria-label='Brush radius']`
+                ) ||
+                gradioApp().querySelector(
+                    `${elemId} button[aria-label="Use brush"]`
+                );
+
+            if (input) {
+                input.click();
+                if (!withoutValue) {
+                    const maxValue =
+                        parseFloat(input.getAttribute("max")) || 100;
+                    const changeAmount = maxValue * (percentage / 100);
+                    const newValue =
+                        parseFloat(input.value) +
+                        (deltaY > 0 ? -changeAmount : changeAmount);
+                    input.value = Math.min(Math.max(newValue, 0), maxValue);
+                    input.dispatchEvent(new Event("change"));
+                }
+            }
+        }
+
+        // Reset zoom when uploading a new image
+        const fileInput = gradioApp().querySelector(
+            `${elemId} input[type="file"][accept="image/*"].svelte-116rqfv`
+        );
+        fileInput.addEventListener("click", resetZoom);
+
+        // Update the zoom level and pan position of the target element based on the values of the zoomLevel, panX and panY variables
+        function updateZoom(newZoomLevel, mouseX, mouseY) {
+            newZoomLevel = Math.max(0.1, Math.min(newZoomLevel, 15));
+
+            elemData[elemId].panX +=
+                mouseX - (mouseX * newZoomLevel) / elemData[elemId].zoomLevel;
+            elemData[elemId].panY +=
+                mouseY - (mouseY * newZoomLevel) / elemData[elemId].zoomLevel;
+
+            targetElement.style.transformOrigin = "0 0";
+            targetElement.style.transform = `translate(${elemData[elemId].panX}px, ${elemData[elemId].panY}px) scale(${newZoomLevel})`;
+
+            toggleOverlap("on");
+            if (isExtension) {
+                targetElement.style.overflow = "visible";
+            }
+
+            return newZoomLevel;
+        }
+
+        // Change the zoom level based on user interaction
+        function changeZoomLevel(operation, e) {
+            if (isModifierKey(e, hotkeysConfig.canvas_hotkey_zoom)) {
+                e.preventDefault();
+
+                let zoomPosX, zoomPosY;
+                let delta = 0.2;
+                if (elemData[elemId].zoomLevel > 7) {
+                    delta = 0.9;
+                } else if (elemData[elemId].zoomLevel > 2) {
+                    delta = 0.6;
+                }
+
+                zoomPosX = e.clientX;
+                zoomPosY = e.clientY;
+
+                fullScreenMode = false;
+                elemData[elemId].zoomLevel = updateZoom(
+                    elemData[elemId].zoomLevel +
+                    (operation === "+" ? delta : -delta),
+                    zoomPosX - targetElement.getBoundingClientRect().left,
+                    zoomPosY - targetElement.getBoundingClientRect().top
+                );
+
+                targetElement.isZoomed = true;
+            }
+        }
+
+        /**
+         * This function fits the target element to the screen by calculating
+         * the required scale and offsets. It also updates the global variables
+         * zoomLevel, panX, and panY to reflect the new state.
+         */
+
+        function fitToElement() {
+            //Reset Zoom
+            targetElement.style.transform = `translate(${0}px, ${0}px) scale(${1})`;
+
+            let parentElement;
+
+            if (isExtension) {
+                parentElement = targetElement.closest('[id^="component-"]');
+            } else {
+                parentElement = targetElement.parentElement;
+            }
+
+
+            // Get element and screen dimensions
+            const elementWidth = targetElement.offsetWidth;
+            const elementHeight = targetElement.offsetHeight;
+
+            const screenWidth = parentElement.clientWidth;
+            const screenHeight = parentElement.clientHeight;
+
+            // Get element's coordinates relative to the parent element
+            const elementRect = targetElement.getBoundingClientRect();
+            const parentRect = parentElement.getBoundingClientRect();
+            const elementX = elementRect.x - parentRect.x;
+
+            // Calculate scale and offsets
+            const scaleX = screenWidth / elementWidth;
+            const scaleY = screenHeight / elementHeight;
+            const scale = Math.min(scaleX, scaleY);
+
+            const transformOrigin =
+                window.getComputedStyle(targetElement).transformOrigin;
+            const [originX, originY] = transformOrigin.split(" ");
+            const originXValue = parseFloat(originX);
+            const originYValue = parseFloat(originY);
+
+            const offsetX =
+                (screenWidth - elementWidth * scale) / 2 -
+                originXValue * (1 - scale);
+            const offsetY =
+                (screenHeight - elementHeight * scale) / 2.5 -
+                originYValue * (1 - scale);
+
+            // Apply scale and offsets to the element
+            targetElement.style.transform = `translate(${offsetX}px, ${offsetY}px) scale(${scale})`;
+
+            // Update global variables
+            elemData[elemId].zoomLevel = scale;
+            elemData[elemId].panX = offsetX;
+            elemData[elemId].panY = offsetY;
+
+            fullScreenMode = false;
+            toggleOverlap("off");
+        }
+
+        /**
+         * This function fits the target element to the screen by calculating
+         * the required scale and offsets. It also updates the global variables
+         * zoomLevel, panX, and panY to reflect the new state.
+         */
+
+        // Fullscreen mode
+        function fitToScreen() {
+            const canvas = gradioApp().querySelector(
+                `${elemId} canvas[key="interface"]`
+            );
+
+            if (!canvas) return;
+
+            if (canvas.offsetWidth > 862 || isExtension) {
+                targetElement.style.width = (canvas.offsetWidth + 2) + "px";
+            }
+
+            if (isExtension) {
+                targetElement.style.overflow = "visible";
+            }
+
+            if (fullScreenMode) {
+                resetZoom();
+                fullScreenMode = false;
+                return;
+            }
+
+            //Reset Zoom
+            targetElement.style.transform = `translate(${0}px, ${0}px) scale(${1})`;
+
+            // Get scrollbar width to right-align the image
+            const scrollbarWidth =
+                window.innerWidth - document.documentElement.clientWidth;
+
+            // Get element and screen dimensions
+            const elementWidth = targetElement.offsetWidth;
+            const elementHeight = targetElement.offsetHeight;
+            const screenWidth = window.innerWidth - scrollbarWidth;
+            const screenHeight = window.innerHeight;
+
+            // Get element's coordinates relative to the page
+            const elementRect = targetElement.getBoundingClientRect();
+            const elementY = elementRect.y;
+            const elementX = elementRect.x;
+
+            // Calculate scale and offsets
+            const scaleX = screenWidth / elementWidth;
+            const scaleY = screenHeight / elementHeight;
+            const scale = Math.min(scaleX, scaleY);
+
+            // Get the current transformOrigin
+            const computedStyle = window.getComputedStyle(targetElement);
+            const transformOrigin = computedStyle.transformOrigin;
+            const [originX, originY] = transformOrigin.split(" ");
+            const originXValue = parseFloat(originX);
+            const originYValue = parseFloat(originY);
+
+            // Calculate offsets with respect to the transformOrigin
+            const offsetX =
+                (screenWidth - elementWidth * scale) / 2 -
+                elementX -
+                originXValue * (1 - scale);
+            const offsetY =
+                (screenHeight - elementHeight * scale) / 2 -
+                elementY -
+                originYValue * (1 - scale);
+
+            // Apply scale and offsets to the element
+            targetElement.style.transform = `translate(${offsetX}px, ${offsetY}px) scale(${scale})`;
+
+            // Update global variables
+            elemData[elemId].zoomLevel = scale;
+            elemData[elemId].panX = offsetX;
+            elemData[elemId].panY = offsetY;
+
+            fullScreenMode = true;
+            toggleOverlap("on");
+        }
+
+        // Handle keydown events
+        function handleKeyDown(event) {
+            // Disable key locks to make pasting from the buffer work correctly
+            if ((event.ctrlKey && event.code === 'KeyV') || (event.ctrlKey && event.code === 'KeyC') || event.code === "F5") {
+                return;
+            }
+
+            // before activating shortcut, ensure user is not actively typing in an input field
+            if (!hotkeysConfig.canvas_blur_prompt) {
+                if (event.target.nodeName === 'TEXTAREA' || event.target.nodeName === 'INPUT') {
+                    return;
+                }
+            }
+
+
+            const hotkeyActions = {
+                [hotkeysConfig.canvas_hotkey_reset]: resetZoom,
+                [hotkeysConfig.canvas_hotkey_overlap]: toggleOverlap,
+                [hotkeysConfig.canvas_hotkey_fullscreen]: fitToScreen,
+                [hotkeysConfig.canvas_hotkey_shrink_brush]: () => adjustBrushSize(elemId, 10),
+                [hotkeysConfig.canvas_hotkey_grow_brush]: () => adjustBrushSize(elemId, -10)
+            };
+
+            const action = hotkeyActions[event.code];
+            if (action) {
+                event.preventDefault();
+                action(event);
+            }
+
+            if (
+                isModifierKey(event, hotkeysConfig.canvas_hotkey_zoom) ||
+                isModifierKey(event, hotkeysConfig.canvas_hotkey_adjust)
+            ) {
+                event.preventDefault();
+            }
+        }
+
+        // Get Mouse position
+        function getMousePosition(e) {
+            mouseX = e.offsetX;
+            mouseY = e.offsetY;
+        }
+
+        // Simulation of the function to put a long image into the screen.
+        // We detect if an image has a scroll bar or not, make a fullscreen to reveal the image, then reduce it to fit into the element.
+        // We hide the image and show it to the user when it is ready.
+
+        targetElement.isExpanded = false;
+        function autoExpand() {
+            const canvas = document.querySelector(`${elemId} canvas[key="interface"]`);
+            if (canvas) {
+                if (hasHorizontalScrollbar(targetElement) && targetElement.isExpanded === false) {
+                    targetElement.style.visibility = "hidden";
+                    setTimeout(() => {
+                        fitToScreen();
+                        resetZoom();
+                        targetElement.style.visibility = "visible";
+                        targetElement.isExpanded = true;
+                    }, 10);
+                }
+            }
+        }
+
+        targetElement.addEventListener("mousemove", getMousePosition);
+
+        //observers
+        // Creating an observer with a callback function to handle DOM changes
+        const observer = new MutationObserver((mutationsList, observer) => {
+            for (let mutation of mutationsList) {
+                // If the style attribute of the canvas has changed, by observation it happens only when the picture changes
+                if (mutation.type === 'attributes' && mutation.attributeName === 'style' &&
+                    mutation.target.tagName.toLowerCase() === 'canvas') {
+                    targetElement.isExpanded = false;
+                    setTimeout(resetZoom, 10);
+                }
+            }
+        });
+
+        // Apply auto expand if enabled
+        if (hotkeysConfig.canvas_auto_expand) {
+            targetElement.addEventListener("mousemove", autoExpand);
+            // Set up an observer to track attribute changes
+            observer.observe(targetElement, {attributes: true, childList: true, subtree: true});
+        }
+
+        // Handle events only inside the targetElement
+        let isKeyDownHandlerAttached = false;
+
+        function handleMouseMove() {
+            if (!isKeyDownHandlerAttached) {
+                document.addEventListener("keydown", handleKeyDown);
+                isKeyDownHandlerAttached = true;
+
+                activeElement = elemId;
+            }
+        }
+
+        function handleMouseLeave() {
+            if (isKeyDownHandlerAttached) {
+                document.removeEventListener("keydown", handleKeyDown);
+                isKeyDownHandlerAttached = false;
+
+                activeElement = null;
+            }
+        }
+
+        // Add mouse event handlers
+        targetElement.addEventListener("mousemove", handleMouseMove);
+        targetElement.addEventListener("mouseleave", handleMouseLeave);
+
+        // Reset zoom when click on another tab
+        elements.img2imgTabs.addEventListener("click", resetZoom);
+        elements.img2imgTabs.addEventListener("click", () => {
+            // targetElement.style.width = "";
+            if (parseInt(targetElement.style.width) > 865) {
+                setTimeout(fitToElement, 0);
+            }
+        });
+
+        targetElement.addEventListener("wheel", e => {
+            // change zoom level
+            const operation = e.deltaY > 0 ? "-" : "+";
+            changeZoomLevel(operation, e);
+
+            // Handle brush size adjustment with ctrl key pressed
+            if (isModifierKey(e, hotkeysConfig.canvas_hotkey_adjust)) {
+                e.preventDefault();
+
+                // Increase or decrease brush size based on scroll direction
+                adjustBrushSize(elemId, e.deltaY);
+            }
+        });
+
+        // Handle the move event for pan functionality. Updates the panX and panY variables and applies the new transform to the target element.
+        function handleMoveKeyDown(e) {
+
+            // Disable key locks to make pasting from the buffer work correctly
+            if ((e.ctrlKey && e.code === 'KeyV') || (e.ctrlKey && event.code === 'KeyC') || e.code === "F5") {
+                return;
+            }
+
+            // before activating shortcut, ensure user is not actively typing in an input field
+            if (!hotkeysConfig.canvas_blur_prompt) {
+                if (e.target.nodeName === 'TEXTAREA' || e.target.nodeName === 'INPUT') {
+                    return;
+                }
+            }
+
+
+            if (e.code === hotkeysConfig.canvas_hotkey_move) {
+                if (!e.ctrlKey && !e.metaKey && isKeyDownHandlerAttached) {
+                    e.preventDefault();
+                    document.activeElement.blur();
+                    isMoving = true;
+                }
+            }
+        }
+
+        function handleMoveKeyUp(e) {
+            if (e.code === hotkeysConfig.canvas_hotkey_move) {
+                isMoving = false;
+            }
+        }
+
+        document.addEventListener("keydown", handleMoveKeyDown);
+        document.addEventListener("keyup", handleMoveKeyUp);
+
+        // Detect zoom level and update the pan speed.
+        function updatePanPosition(movementX, movementY) {
+            let panSpeed = 2;
+
+            if (elemData[elemId].zoomLevel > 8) {
+                panSpeed = 3.5;
+            }
+
+            elemData[elemId].panX += movementX * panSpeed;
+            elemData[elemId].panY += movementY * panSpeed;
+
+            // Delayed redraw of an element
+            requestAnimationFrame(() => {
+                targetElement.style.transform = `translate(${elemData[elemId].panX}px, ${elemData[elemId].panY}px) scale(${elemData[elemId].zoomLevel})`;
+                toggleOverlap("on");
+            });
+        }
+
+        function handleMoveByKey(e) {
+            if (isMoving && elemId === activeElement) {
+                updatePanPosition(e.movementX, e.movementY);
+                targetElement.style.pointerEvents = "none";
+
+                if (isExtension) {
+                    targetElement.style.overflow = "visible";
+                }
+
+            } else {
+                targetElement.style.pointerEvents = "auto";
+            }
+        }
+
+        // Prevents sticking to the mouse
+        window.onblur = function() {
+            isMoving = false;
+        };
+
+        // Checks for extension
+        function checkForOutBox() {
+            const parentElement = targetElement.closest('[id^="component-"]');
+            if (parentElement.offsetWidth < targetElement.offsetWidth && !targetElement.isExpanded) {
+                resetZoom();
+                targetElement.isExpanded = true;
+            }
+
+            if (parentElement.offsetWidth < targetElement.offsetWidth && elemData[elemId].zoomLevel == 1) {
+                resetZoom();
+            }
+
+            if (parentElement.offsetWidth < targetElement.offsetWidth && targetElement.offsetWidth * elemData[elemId].zoomLevel > parentElement.offsetWidth && elemData[elemId].zoomLevel < 1 && !targetElement.isZoomed) {
+                resetZoom();
+            }
+        }
+
+        if (isExtension) {
+            targetElement.addEventListener("mousemove", checkForOutBox);
+        }
+
+
+        window.addEventListener('resize', (e) => {
+            resetZoom();
+
+            if (isExtension) {
+                targetElement.isExpanded = false;
+                targetElement.isZoomed = false;
+            }
+        });
+
+        gradioApp().addEventListener("mousemove", handleMoveByKey);
+
+
+    }
+
+    applyZoomAndPan(elementIDs.sketch, false);
+    applyZoomAndPan(elementIDs.inpaint, false);
+    applyZoomAndPan(elementIDs.inpaintSketch, false);
+
+    // Make the function global so that other extensions can take advantage of this solution
+    const applyZoomAndPanIntegration = async(id, elementIDs) => {
+        const mainEl = document.querySelector(id);
+        if (id.toLocaleLowerCase() === "none") {
+            for (const elementID of elementIDs) {
+                const el = await waitForElement(elementID);
+                if (!el) break;
+                applyZoomAndPan(elementID);
+            }
+            return;
+        }
+
+        if (!mainEl) return;
+        mainEl.addEventListener("click", async() => {
+            for (const elementID of elementIDs) {
+                const el = await waitForElement(elementID);
+                if (!el) break;
+                applyZoomAndPan(elementID);
+            }
+        }, {once: true});
+    };
+
+    window.applyZoomAndPan = applyZoomAndPan; // Only 1 elements, argument elementID, for example applyZoomAndPan("#txt2img_controlnet_ControlNet_input_image")
+
+    window.applyZoomAndPanIntegration = applyZoomAndPanIntegration; // for any extension
+
+    /*
+        The function `applyZoomAndPanIntegration` takes two arguments:
+
+        1. `id`: A string identifier for the element to which zoom and pan functionality will be applied on click.
+        If the `id` value is "none", the functionality will be applied to all elements specified in the second argument without a click event.
+
+        2. `elementIDs`: An array of string identifiers for elements. Zoom and pan functionality will be applied to each of these elements on click of the element specified by the first argument.
+        If "none" is specified in the first argument, the functionality will be applied to each of these elements without a click event.
+
+        Example usage:
+        applyZoomAndPanIntegration("#txt2img_controlnet", ["#txt2img_controlnet_ControlNet_input_image"]);
+        In this example, zoom and pan functionality will be applied to the element with the identifier "txt2img_controlnet_ControlNet_input_image" upon clicking the element with the identifier "txt2img_controlnet".
+    */
+
+    // More examples
+    // Add integration with ControlNet txt2img One TAB
+    // applyZoomAndPanIntegration("#txt2img_controlnet", ["#txt2img_controlnet_ControlNet_input_image"]);
+
+    // Add integration with ControlNet txt2img Tabs
+    // applyZoomAndPanIntegration("#txt2img_controlnet",Array.from({ length: 10 }, (_, i) => `#txt2img_controlnet_ControlNet-${i}_input_image`));
+
+    // Add integration with Inpaint Anything
+    // applyZoomAndPanIntegration("None", ["#ia_sam_image", "#ia_sel_mask"]);
+});

extensions-builtin/canvas-zoom-and-pan/scripts/hotkey_config.py

@@ -0,0 +1,17 @@
+import gradio as gr
+from modules import shared
+
+shared.options_templates.update(shared.options_section(('canvas_hotkey', "Canvas Hotkeys"), {
+    "canvas_hotkey_zoom": shared.OptionInfo("Alt", "Zoom canvas", gr.Radio, {"choices": ["Shift","Ctrl", "Alt"]}).info("If you choose 'Shift' you cannot scroll horizontally, 'Alt' can cause a little trouble in firefox"),
+    "canvas_hotkey_adjust": shared.OptionInfo("Ctrl", "Adjust brush size", gr.Radio, {"choices": ["Shift","Ctrl", "Alt"]}).info("If you choose 'Shift' you cannot scroll horizontally, 'Alt' can cause a little trouble in firefox"),
+    "canvas_hotkey_shrink_brush": shared.OptionInfo("Q", "Shrink the brush size"),
+    "canvas_hotkey_grow_brush": shared.OptionInfo("W", "Enlarge the brush size"),
+    "canvas_hotkey_move": shared.OptionInfo("F", "Moving the canvas").info("To work correctly in firefox, turn off 'Automatically search the page text when typing' in the browser settings"),
+    "canvas_hotkey_fullscreen": shared.OptionInfo("S", "Fullscreen Mode, maximizes the picture so that it fits into the screen and stretches it to its full width "),
+    "canvas_hotkey_reset": shared.OptionInfo("R", "Reset zoom and canvas positon"),
+    "canvas_hotkey_overlap": shared.OptionInfo("O", "Toggle overlap").info("Technical button, neededs for testing"),
+    "canvas_show_tooltip": shared.OptionInfo(True, "Enable tooltip on the canvas"),
+    "canvas_auto_expand": shared.OptionInfo(True, "Automatically expands an image that does not fit completely in the canvas area, similar to manually pressing the S and R buttons"),
+    "canvas_blur_prompt": shared.OptionInfo(False, "Take the focus off the prompt when working with a canvas"),
+    "canvas_disabled_functions": shared.OptionInfo(["Overlap"], "Disable function that you don't use", gr.CheckboxGroup, {"choices": ["Zoom","Adjust brush size","Hotkey enlarge brush","Hotkey shrink brush","Moving canvas","Fullscreen","Reset Zoom","Overlap"]}),
+}))

extensions-builtin/canvas-zoom-and-pan/style.css

@@ -0,0 +1,66 @@
+.canvas-tooltip-info {
+  position: absolute;
+  top: 10px;
+  left: 10px;
+  cursor: help;
+  background-color: rgba(0, 0, 0, 0.3);
+  width: 20px;
+  height: 20px;
+  border-radius: 50%;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  flex-direction: column; 
+
+  z-index: 100;
+}
+
+.canvas-tooltip-info::after {
+  content: '';
+  display: block;
+  width: 2px;
+  height: 7px;
+  background-color: white;
+  margin-top: 2px; 
+}
+
+.canvas-tooltip-info::before {
+  content: '';
+  display: block;
+  width: 2px;
+  height: 2px;
+  background-color: white;
+}
+
+.canvas-tooltip-content {
+  display: none;
+  background-color: #f9f9f9; 
+  color: #333; 
+  border: 1px solid #ddd;
+  padding: 15px; 
+  position: absolute;
+  top: 40px;
+  left: 10px;
+  width: 250px;
+  font-size: 16px;
+  opacity: 0;
+  border-radius: 8px;
+  box-shadow: 0px 8px 16px 0px rgba(0,0,0,0.2); 
+
+  z-index: 100;
+}
+
+.canvas-tooltip:hover .canvas-tooltip-content {
+  display: block;
+  animation: fadeIn 0.5s; 
+  opacity: 1;
+}
+
+@keyframes fadeIn {
+  from {opacity: 0;}
+  to {opacity: 1;}
+}
+
+.styler {
+  overflow:inherit !important;
+}

extensions-builtin/extra-options-section/scripts/extra_options_section.py

@@ -0,0 +1,78 @@
+import math
+
+import gradio as gr
+from modules import scripts, shared, ui_components, ui_settings, infotext_utils
+from modules.ui_components import FormColumn
+
+
+class ExtraOptionsSection(scripts.Script):
+    section = "extra_options"
+
+    def __init__(self):
+        self.comps = None
+        self.setting_names = None
+
+    def title(self):
+        return "Extra options"
+
+    def show(self, is_img2img):
+        return scripts.AlwaysVisible
+
+    def ui(self, is_img2img):
+        self.comps = []
+        self.setting_names = []
+        self.infotext_fields = []
+        extra_options = shared.opts.extra_options_img2img if is_img2img else shared.opts.extra_options_txt2img
+        elem_id_tabname = "extra_options_" + ("img2img" if is_img2img else "txt2img")
+
+        mapping = {k: v for v, k in infotext_utils.infotext_to_setting_name_mapping}
+
+        with gr.Blocks() as interface:
+            with gr.Accordion("Options", open=False, elem_id=elem_id_tabname) if shared.opts.extra_options_accordion and extra_options else gr.Group(elem_id=elem_id_tabname):
+
+                row_count = math.ceil(len(extra_options) / shared.opts.extra_options_cols)
+
+                for row in range(row_count):
+                    with gr.Row():
+                        for col in range(shared.opts.extra_options_cols):
+                            index = row * shared.opts.extra_options_cols + col
+                            if index >= len(extra_options):
+                                break
+
+                            setting_name = extra_options[index]
+
+                            with FormColumn():
+                                comp = ui_settings.create_setting_component(setting_name)
+
+                            self.comps.append(comp)
+                            self.setting_names.append(setting_name)
+
+                            setting_infotext_name = mapping.get(setting_name)
+                            if setting_infotext_name is not None:
+                                self.infotext_fields.append((comp, setting_infotext_name))
+
+        def get_settings_values():
+            res = [ui_settings.get_value_for_setting(key) for key in self.setting_names]
+            return res[0] if len(res) == 1 else res
+
+        interface.load(fn=get_settings_values, inputs=[], outputs=self.comps, queue=False, show_progress=False)
+
+        return self.comps
+
+    def before_process(self, p, *args):
+        for name, value in zip(self.setting_names, args):
+            if name not in p.override_settings:
+                p.override_settings[name] = value
+
+
+shared.options_templates.update(shared.options_section(('settings_in_ui', "Settings in UI", "ui"), {
+    "settings_in_ui": shared.OptionHTML("""
+This page allows you to add some settings to the main interface of txt2img and img2img tabs.
+"""),
+    "extra_options_txt2img": shared.OptionInfo([], "Settings for txt2img", ui_components.DropdownMulti, lambda: {"choices": list(shared.opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that also appear in txt2img interfaces").needs_reload_ui(),
+    "extra_options_img2img": shared.OptionInfo([], "Settings for img2img", ui_components.DropdownMulti, lambda: {"choices": list(shared.opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that also appear in img2img interfaces").needs_reload_ui(),
+    "extra_options_cols": shared.OptionInfo(1, "Number of columns for added settings", gr.Slider, {"step": 1, "minimum": 1, "maximum": 20}).info("displayed amount will depend on the actual browser window width").needs_reload_ui(),
+    "extra_options_accordion": shared.OptionInfo(False, "Place added settings into an accordion").needs_reload_ui()
+}))
+
+

extensions-builtin/forge_legacy_preprocessors/.gitignore

@@ -0,0 +1,185 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea
+*.pt
+*.pth
+*.ckpt
+*.bin
+*.safetensors
+
+# Editor setting metadata
+.idea/
+.vscode/
+detected_maps/
+annotator/downloads/
+
+# test results and expectations
+web_tests/results/
+web_tests/expectations/
+tests/web_api/full_coverage/results/
+tests/web_api/full_coverage/expectations/
+
+*_diff.png
+
+# Presets
+presets/
+
+# Ignore existing dir of hand refiner if exists.
+annotator/hand_refiner_portable