ComfyUI-Pixelate

ComfyUI-Pixelate
★ 5

像素化色彩量化下采样抖动处理
在ComfyUI中实现像素化与高质量下采样与颜色量化，支持多种缩放算法、色彩模式、抖动与调色板生成，适合生成像素艺术和调色处理。
💡 在ComfyUI流程中将照片转为像素艺术并保留色彩风格
🍴 1 Forks💻 Python🔄 2024-11-26
📦
网盘下载
复制链接后前往夸克网盘下载
https://pan.quark.cn/s/a9fb3a59e10c
ComfyUIPixelate
📄 README

ComfyUIPixelate

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

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

  • Add the “Pixelate” node to your ComfyUI workflow
  • Connect an image input
  • 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

  • Original concept based on sd-webui-pixelart