KRForge

Portable Stable Diffusion build for exterior architectural visualization, based on Forge Neo (AGPL-3.0).

Install

  1. Create an empty folder without spaces in the path, for example C:\KRForge.
  2. Put Install_KRForge.bat into it and run it.
  3. After the install, open the Model Manager tab and install Dark Beast (BF16 or FP8) or Z-Image Turbo.

Start the build with cs_1_lite.bat (simple UI) or cs_2_full.bat (full UI). If some base model did not download during the install, run download_models.bat (it skips files that are already present). To update the build, run sdw\._INSTALL_LAST_PATCH.bat.

Files in this repository

File What it is
Install_KRForge.bat installer
soft.zip tools for the installer: wget, lz4, tar
sys.tar.lz4 portable Python environment
sdw.tar.lz4 Forge Neo with settings and extensions (no models)
setup/setup.bat install steps (base models via download_models.bat inside sdw.tar.lz4)
setup/setup_patch.bat current patch (updates)
models/model-manager.json model list for the Model Manager tab
models/files/ two model files that have no direct download from their authors

Models and their sources

Models are not part of the archives. Base models are downloaded from their authors by download_models.bat during the install; Dark Beast and Z-Image Turbo are installed in the Model Manager tab. Each model is under its own license; check it before commercial use.

Model Source License
Dark Beast BlitZ ZiT (BF16, FP8) GuangyuanSD/REDCraft-DarkBeast-Z-Image-TURBO (author AiMetatron) Apache-2.0
Z-Image Turbo, Qwen3 4B text encoder, VAE Comfy-Org/z_image_turbo Apache-2.0 (original: Tongyi-MAI/Z-Image-Turbo)
epiCPhotoGasm Last Unicorn Civitai, epinikion see model page
ControlNet 1.1 SD1.5 (canny, depth, inpaint, lineart, tile) comfyanonymous/ControlNet-v1-1_fp16_safetensors OpenRAIL (original: lllyasviel/ControlNet-v1-1)
IP-Adapter SD1.5 plus lllyasviel/sd_control_collection Apache-2.0 (original: h94/IP-Adapter)
4x-UltraSharp Kim2091/UltraSharp CC BY-NC-SA 4.0
4x NMKD-Siax 200k gemasai/4x_NMKD-Siax_200k see source
Advanced Enhancer LoRA Civitai see model page
Detail Slider LoRA v4 Civitai, ostris see model page
Depth Anything ControlNet (models/files) LiheYoung/Depth-Anything see source
Realistic Snapshot ZIT v5 (models/files, LoRA strength 0.6-0.7) Civitai, MonkeyForever redistribution allowed by the author

ControlNet preprocessors, CLIP-ViT-H and BLIP are downloaded automatically by Forge on first use.

Source code

Forge Neo is licensed under AGPL-3.0. The full source code of this build, including our changes (Model Manager, launch scripts), is inside sdw.tar.lz4 (folder sdw, with git history).

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support