Conflict With Dynamic Prompt (Wildcards manager tab)

README.md

@@ -3,8 +3,10 @@ This extension is an extension of the built-in Composable Diffusion.
 This allows you to determine the region of the latent space that reflects your subprompts.
 
 ## How to use
+![20230303.png](./screenshots/20230303.png)
 ![20230213.png](./screenshots/20230213.png)
 
+
 ### Enabled
 The effect of Latent Couple appears only when Enabled is checked.
 
@@ -41,11 +43,34 @@ outputs
 - end at step=4 https://imgur.com/a1kyvhX
 - end at step=0 https://imgur.com/yhGF7g8
 
-## Old prerequisite
-This extension need to apply cfg_denoised_callback-ea9bd9fc.patch (as of Feb 5, 2023 origin/HEAD commit ea9bd9fc).
+
+## ~~Prerequisite for prompt pasting~~
+## ~~Prerequisite for gradio Image and Sketch component bug fix~~
+This fix is no longer suitable for latest webui commit at 22bcc7be, with gradio dependency upgraded to 3.23. 
+
+I'll keep the fix here for people still using older versions of webui.
+
+Activate your venv in webui root directory
+
+For Windows, in cmd
+```
+venv\Scripts\activate.bat
 ```
-git apply --ignore-whitespace extensions/stable-diffusion-webui-two-shot/cfg_denoised_callback-ea9bd9fc.patch
+For Linux
 ```
+source venv/bin/activate
+```
+Then, install wheel distribution with bugfix applied
+```
+pip install --force-reinstall --no-deps extensions/stable-diffusion-webui-two-shot/gradio-3.16.2-py3-none-any.whl
+```
+For bugfix related modifications, see https://github.com/ashen-sensored/gradio/tree/3.16.2
+
+
+## Issues
+- ~~The extension's mask color sketching function does not work well with chrome(extreme stuttering) due to gradio's Image component bug.~~ Please keep the browser scaling at 100% while creating blank canvas to avoid the bug. 
+See prerequisite above. The fix is no longer suitable for latest webui version at 22bcc7be, with gradio dependency upgraded to 3.23. 
+
 
 ## Credits
 - two shot diffusion.ipynb https://colab.research.google.com/drive/1UdElpQfKFjY5luch9v_LlmSdH7AmeiDe?usp=sharing

scripts/sketch_helper.py

@@ -0,0 +1,99 @@
+import numpy as np
+import cv2
+import base64
+
+
+def count_high_freq_colors(image):
+    im = image.getcolors(maxcolors=1024*1024)
+    sorted_colors = sorted(im, key=lambda x: x[0], reverse=True)
+
+    freqs = [c[0] for c in sorted_colors]
+    mean_freq = sum(freqs) / len(freqs)
+
+    high_freq_colors = [c for c in sorted_colors if c[0] > max(2, mean_freq*1.25)]
+    return high_freq_colors
+
+def get_high_freq_colors(image, similarity_threshold=30):
+    image_copy = image.copy()
+    high_freq_colors = count_high_freq_colors(image)
+    # Check for similar colors and replace the lower frequency color with the higher frequency color in the image
+    for i, (freq1, color1) in enumerate(high_freq_colors):
+        for j, (freq2, color2) in enumerate(high_freq_colors):      
+            if (color_distance(color1, color2) < similarity_threshold) or (color_distance(color1, opaque_color_on_white(color2, 0.5)) < 5):
+              if(freq2 > freq1):
+                replace_color(image_copy, color1, color2)
+
+    high_freq_colors = count_high_freq_colors(image_copy)
+    print(high_freq_colors)
+    return [high_freq_colors, image_copy]
+
+def color_quantization(image, color_frequency_list):
+    # Convert the color frequency list to a set of unique colors
+    unique_colors = set([color for _, color in color_frequency_list])
+
+    # Create a mask for the image with True where the color is in the unique colors set
+    mask = np.any(np.all(image.reshape(-1, 1, 3) == np.array(list(unique_colors)), axis=2), axis=1).reshape(image.shape[:2])
+
+    # Create a new image with all pixels set to white
+    new_image = np.full_like(image, 255)
+
+    # Copy the pixels from the original image that have a color in the color frequency list
+    new_image[mask] = image[mask]
+    return new_image
+
+
+def create_binary_mask(img_arr, target_color):
+    # Create mask of pixels with target color
+    mask = np.all(img_arr == target_color, axis=-1)
+
+    # Convert mask to binary matrix
+    binary_matrix = mask.astype(int)
+
+    return binary_matrix
+
+
+def create_binary_matrix_base64(img_arr, target_color):
+    # Create mask of pixels with target color
+    mask = np.all(img_arr == target_color, axis=-1)
+
+    # Convert mask to binary matrix
+    binary_matrix = mask.astype(int)
+    from datetime import datetime
+    binary_file_name = f'mask-{datetime.now().timestamp()}.png'
+    _, im_arr = cv2.imencode('.png', binary_matrix * 255)
+
+    im_bytes = im_arr.tobytes()
+    im_b64 = base64.b64encode(im_bytes).decode('ascii')
+    # binary_matrix = torch.from_numpy(binary_matrix).unsqueeze(0).unsqueeze(0)
+    return mask, im_b64
+
+def create_binary_matrix_img(img_arr, target_color):
+    # Create mask of pixels with target color
+    mask = np.all(img_arr == target_color, axis=-1)
+
+    # Convert mask to binary matrix
+    binary_matrix = mask.astype(int)
+    from datetime import datetime
+    binary_file_name = f'mask-{datetime.now().timestamp()}.png'
+    cv2.imwrite(binary_file_name, binary_matrix * 255)
+
+    #binary_matrix = torch.from_numpy(binary_matrix).unsqueeze(0).unsqueeze(0)
+    return binary_file_name
+
+def color_distance(color1, color2):
+    return sum((a - b) ** 2 for a, b in zip(color1, color2)) ** 0.5
+
+def replace_color(image, old_color, new_color):
+    pixels = image.load()
+    width, height = image.size
+    for x in range(width):
+        for y in range(height):
+            if pixels[x, y] == old_color:
+                pixels[x, y] = new_color
+
+def opaque_color_on_white(color, a):
+    r, g, b = color
+    opaque_red = int((1 - a) * 255 + a * r)
+    opaque_green = int((1 - a) * 255 + a * g)
+    opaque_blue = int((1 - a) * 255 + a * b)
+    return (opaque_red, opaque_green, opaque_blue)