first commit, magicboard on hf

README.md

@@ -1,13 +1,32 @@
----
-title: MagicBoard
-emoji: 🐨
-colorFrom: red
-colorTo: green
-sdk: streamlit
-sdk_version: 1.10.0
-app_file: app.py
-pinned: false
-license: mit
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Magic Board
+Magic Board is a PoC using stable diffusion to generate out-of-mind images starting from a sketch.
+
+Thank to streamlit-drawable-canvas package it is super simple create the perfect environment to play with stable diffusion, starting form our quick draw.
+Select the number of steps and the strength to generate the final image.
+
+I like minimalism, so for this experiments I try to use a low-code approach. Also lowest time as possible ;)
+
+## Some technicalities
+
+The Stable Diffusion model is the latest provided by the diffusers library.
+
+If you want to use it, to avoid local checkpoint, just get an access token from hugging face, follow [these](https://huggingface.co/CompVis/stable-diffusion-v1-4) instruction to get access.
+Then copy and paste the Token in [config.py](./config.py).
+
+Install the requirements and you should be able to run:
+```
+streamlit run main.py
+```
+
+This version has also the multi-language support, at the moment just English and Italian.
+
+![Screen dashboard](./magic.png)
+
+
+The final result is :
+<p style="text-align:center;"><img src="./result.png" width="200"/></p>
+
+
+NOTE:
+
+This implementation use a `torch.float16` due to low VRAM availability. Strongly suggest to set it `torch.float32` to speed up the computation. [Here](./utils.py)

locales/base.pot

@@ -0,0 +1,56 @@
+# SOME DESCRIPTIVE TITLE.
+# Copyright (C) YEAR ORGANIZATION
+# FIRST AUTHOR <EMAIL@ADDRESS>, YEAR.
+#
+msgid ""
+msgstr ""
+"Project-Id-Version: PACKAGE VERSION\n"
+"POT-Creation-Date: 2022-09-06 00:56+0200\n"
+"PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\n"
+"Last-Translator: FULL NAME <EMAIL@ADDRESS>\n"
+"Language-Team: LANGUAGE <LL@li.org>\n"
+"MIME-Version: 1.0\n"
+"Content-Type: text/plain; charset=UTF-8\n"
+"Content-Transfer-Encoding: 8bit\n"
+"Generated-By: pygettext.py 1.5\n"
+
+#: main.py:10
+msgid "Choose your language"
+msgstr ""
+
+#: main.py:21
+msgid "Drawing tool:"
+msgstr ""
+
+#: main.py:24
+msgid "Stroke width: "
+msgstr ""
+
+#: main.py:25
+msgid "Stroke color hex: "
+msgstr ""
+
+#: main.py:26
+msgid "Background color hex: "
+msgstr ""
+
+#: main.py:27
+msgid "Background image:"
+msgstr ""
+
+#: main.py:28
+msgid "Number of Steps: "
+msgstr ""
+
+#: main.py:29
+msgid "Strength value: "
+msgstr ""
+
+#: main.py:50
+msgid "Prompt to generate your cool image"
+msgstr ""
+
+#: main.py:51
+msgid "write you text here to improve the image with your favourite style"
+msgstr ""
+

locales/it/LC_MESSAGES/base.po

@@ -0,0 +1,56 @@
+# SOME DESCRIPTIVE TITLE.
+# Copyright (C) YEAR ORGANIZATION
+# FIRST AUTHOR <EMAIL@ADDRESS>, YEAR.
+#
+msgid ""
+msgstr ""
+"Project-Id-Version: PACKAGE VERSION\n"
+"POT-Creation-Date: 2022-09-06 00:56+0200\n"
+"PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\n"
+"Last-Translator: FULL NAME <EMAIL@ADDRESS>\n"
+"Language-Team: LANGUAGE <LL@li.org>\n"
+"MIME-Version: 1.0\n"
+"Content-Type: text/plain; charset=UTF-8\n"
+"Content-Transfer-Encoding: 8bit\n"
+"Generated-By: pygettext.py 1.5\n"
+
+#: main.py:10
+msgid "Choose your language"
+msgstr "Seleziona la lingua"
+
+#: main.py:21
+msgid "Drawing tool:"
+msgstr "Strumento da disegno:"
+
+#: main.py:24
+msgid "Stroke width: "
+msgstr "Spessore tratto:"
+
+#: main.py:25
+msgid "Stroke color hex: "
+msgstr "Colore tratto:"
+
+#: main.py:26
+msgid "Background color hex: "
+msgstr "Colore background:"
+
+#: main.py:27
+msgid "Background image:"
+msgstr "Carica un'immagine:"
+
+#: main.py:28
+msgid "Number of Steps: "
+msgstr "Numero di iterazioni:"
+
+#: main.py:29
+msgid "Strength value: "
+msgstr "Attinenza all'originale:"
+
+#: main.py:50
+msgid "Prompt to generate your cool image"
+msgstr "Testo di input per iniziare la magia"
+
+#: main.py:51
+msgid "write you text here to improve the image with your favourite style"
+msgstr "Scrivi qui le indicazioni che trasformeranno la tua immagine"
+

main.py

@@ -0,0 +1,64 @@
+from PIL import Image
+from utils import pipe_image
+import streamlit as st
+from streamlit_drawable_canvas import st_canvas
+import gettext
+import numpy as np
+
+_ = gettext.gettext
+
+language = st.selectbox(_('Choose your language'), ['en', 'it'])
+try:
+  localizator = gettext.translation('base', localedir='locales', languages=[language])
+  localizator.install()
+  _ = localizator.gettext 
+except:
+    pass
+
+st.markdown("<h1 style='text-align:center;'> Magic Board</h1>", unsafe_allow_html=True)
+# Specify canvas parameters in application
+drawing_mode = st.sidebar.selectbox(
+    _("Drawing tool:"), ("freedraw", "line", "rect", "circle", "transform")
+)
+
+stroke_width = st.sidebar.slider(_("Stroke width: "), 1, 25, 3)
+stroke_color = st.sidebar.color_picker(_("Stroke color hex: "))
+bg_color = st.sidebar.color_picker(_("Background color hex: "), "#eee")
+bg_image = st.sidebar.file_uploader(_("Background image:"), type=["png", "jpg"])
+strength = st.sidebar.slider(_("Strength value: "), 0.0, 1.0, 0.1)
+
+# Create a canvas component
+canvas_result = st_canvas(
+    fill_color="rgba(255, 165, 0, 0.3)",  # Fixed fill color with some opacity
+    stroke_width=stroke_width,
+    stroke_color=stroke_color,
+    background_color=bg_color,
+    background_image=Image.open(bg_image) if bg_image else None,
+    update_streamlit=True,
+    height=450,
+    width=680,
+    drawing_mode=drawing_mode,
+    #point_display_radius=point_display_radius if drawing_mode == 'point' else 0,
+    key="canvas",
+)
+
+prompt = st.text_input(_('Prompt to generate your cool image'), 
+               placeholder=_('write you text here to improve the image with your favourite style'))
+
+if st.button('Submit'):
+    img_size = (512, 512)
+    if bg_image:
+        img_back=Image.open(bg_image).convert('RGB').resize(img_size)
+        comp_img = img_back
+    if np.any(canvas_result.image_data):
+        img_draw=Image.fromarray(canvas_result.image_data).convert('RGB').resize(img_size)
+        comp_img = img_draw
+    if bg_image and np.any(canvas_result.image_data):
+        comp_img = Image.new("RGB", img_size)
+        comp_img = Image.blend(img_draw, img_back, alpha=0.5)
+
+    images = pipe_image(prompt=prompt,
+                       init_image=comp_img,
+                       strength=strength)
+    for image in images:
+        st.image(image)

requirements.txt

@@ -0,0 +1,5 @@
+torch
+transformers
+scipy
+ftfy
+git+https://github.com/huggingface/diffusers

utils.py

@@ -0,0 +1,25 @@
+import torch
+from diffusers import StableDiffusionImg2ImgPipeline
+
+import os
+TOKEN=os.environ.get('sd')
+
+torch.cuda.empty_cache()
+device = "cpu"
+
+def pipe_image(prompt,
+               init_image=None,
+               strength=0.6):
+    pipe = StableDiffusionImg2ImgPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", 
+                                                revision="fp16", 
+                                                torch_dtype=torch.float32,
+                                                use_auth_token=TOKEN)
+
+    pipe = pipe.to(device)
+    pipe.enable_attention_slicing()
+    images = pipe(prompt=prompt,
+                    init_image=init_image,
+                    strength=strength,
+                    guidance_scale=7.5,
+                    num_inference_steps=1000).images 
+    return images