Pixydust Quantizer for ComfyUI

Pixydust Quantizer is a custom node extension for ComfyUI that allows for the simplified recreation of tile patterns used in 1990s 16-color PC graphics, offering advanced color quantization and palette optimization features.

• Original

• Pixydust Quantizer

• Regular tool (ImageQuantize)

Features and Concept

• Simplified recreation of tile patterns used in 1990s 16-color PC graphics
• Prioritizes colors used in large areas of the image (e.g., a character's face) for the palette, preventing unnecessary tile patterning
• Threshold settings to adjust the balance between tile patterns and flat colors
• External palette input function to maintain palette consistency in videos or image sequences
• Control over which colors are tile-patterned, which was difficult with traditional methods

The name "Pixydust" combines "pixy" (as in pixel art) and "dust" (to suggest a grainy look), with an added nuance of a clever, tricky effect, like the magic of fairy dust.

Installation

1. Navigate to the custom node directory in ComfyUI:

   cd ComfyUI/custom_nodes/
   

2. Clone this repository:

   git clone https://github.com/yourusername/ComfyUI-Pixydust-Quantizer.git
   

3. Install the required dependencies:

   pip install -r ComfyUI-Pixydust-Quantizer/requirements.txt
   

4. Restart ComfyUI

Usage

PixydustQuantize1 (Primary Color Reduction Node)

This node reduces the colors of an input image using the specified method and maximum number of colors.

Inputs:

• image: Input image
• color_reduction_method: Choose from "Pillow Quantize", "K-Means", or "MedianCut"
• max_colors: Maximum number of colors after reduction

Outputs:

• Reduced Color Image: Image with reduced colors
• Palette Preview: Preview of the color palette used
• Palette Tensor: Color palette in tensor format

PixydustQuantize2 (Palette Optimization Node)

This node optimizes the palette of a color-reduced image and applies dithering. It assigns palette colors based on the proportion of each color used in the image to avoid tile patterning in prominent areas like a character’s face.

Inputs:

• reduced_image: Input image with reduced colors
• fixed_colors: Number of colors in the optimized palette
• reduction_method: Choose from "K-Means" or "MedianCut"
• dither_pattern: Choose from "None", "2x2 Bayer", "4x4 Bayer", or "8x8 Bayer"
• color_distance_threshold: Threshold for using tile patterns
• palette_tensor (optional): External palette input for videos

Outputs:

• Optimized Image: Image with optimized palette and dithering applied
• Color Histogram: Color histogram of the optimized image
• Fixed Palette: Optimized color palette in tensor format

Example Workflow

Here’s a simple workflow example:

1. Load an image
2. Connect the image to the Pixydust Quantize-1
3. Connect the output of the ColorReducerNode to the Pixydust Quantize-2
4. Save or display the resulting optimized image

Note: The processing speed isn't very fast so far. It takes about 10 seconds (depending on the machine's power) to generate a 512x512 pixel image.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Special thanks to the ComfyUI community for providing the tools and creative motivation!