ComfyUI Extension: ComfyUI-Pixelate

Authored by flycarl

Created

Updated

3 stars

a/sd-webui-pixelart are referenced by many webui users, this node is mean to use it in ComfyUI.

Custom Nodes (1)

README

ComfyUIPixelate

sd-webui-pixelart are referenced by many webui users, this node is mean to use it in ComfyUI.

ComfyUIPixelate

Features

  • Downscaling Options: Multiple high-quality scaling algorithms:

    • auto: Automatically selects the best method
    • nearest: Best for preserving exact colors
    • area: Optimal for general downscaling
    • linear: Smooth transitions but may blur
    • cubic: Sharper edges than linear
    • lanczos: High quality with edge preservation

  • Color Processing:

    • RGB color mode
    • Grayscale conversion
    • Binary (Black & White) conversion

  • Advanced Color Quantization:

    • Multiple palette generation methods:
      • auto: Smart method selection based on image size and color count
      • libimagequant: High-quality quantization
      • kmeans: GPU-accelerated when available (falls back to CPU)
      • mediancut: Fast with good quality
      • maxcoverage: Better color distribution
      • fastoctree: Fastest option for large images
      • median_cut: Custom implementation

  • Dithering Support:

    • Floyd-Steinberg dithering for smooth color transitions
    • Simple quantization for clean, sharp results

  • Custom Palette Support:

    • Use reference images to extract palettes
    • Control palette size (2-256 colors)

Usage

  1. Add the "Pixelate" node to your ComfyUI workflow
  2. Connect an image input
  3. Configure parameters:
    • downscale_factor: How much to reduce the image (1-32)
    • scale_mode: Choose scaling algorithm
    • rescale_to_original: Option to restore original size
    • color_mode: RGB/Grayscale/BW
    • colors: Number of colors in output (2-256)
    • quantization_method: Palette generation method
    • dithering: None or Floyd-Steinberg
    • Optional: Connect a palette reference image

Performance Considerations

  • The node automatically selects optimal methods based on image size:
    • Large images (>1M pixels) or many colors (>32): Uses fast octree
    • Medium images (>500K pixels): Uses libimagequant
    • Small images: Uses k-means clustering
  • GPU acceleration for k-means when available
  • Caching for color quantization to improve speed

Credits