Extensions/ComfyUI-Pixelization
ComfyUI Extension

ComfyUI-Pixelization

ComfyUI Node to pixelize images

By DarioFT·Created 7 months ago·Updated 7 months ago· 1
DarioFT/ComfyUI-Pixelization
Nodes1
On cloudLocal install
Categorypostprocessing/Effects
Stars1
Updated7 months ago

Nodes (1)

Pixelization

Pixel art that doesn't look like a resized mess

postprocessing/Effects
Readme

ComfyUI Pixelization

A ComfyUI node that converts images to pixel art using the neural network from the SIGGRAPH Asia paper "Make Your Own Sprites: Aliasing-Aware and Cell-Controllable Pixelization".

Examples

| Original | Pixelized | |----------|-----------| | | |

Installation

ComfyUI Manager (Recommended)

Search for "Pixelization" in ComfyUI Manager and install.

Manual Installation

cd ComfyUI/custom_nodes
git clone https://github.com/DarioFT/ComfyUI-Pixelization
cd ComfyUI-Pixelization
pip install -r requirements.txt

Restart ComfyUI after installation.

Models

Models are downloaded automatically on first use from HuggingFace to:

ComfyUI/models/Pixelization/
├── pixelart_vgg19.pth   (558 MB)
├── 160_net_G_A.pth      (151 MB)
└── alias_net.pth        (35 MB)

Usage

Find the node at: postprocessing/Effects > Pixelization

Parameters

| Parameter | Default | Description | |-----------|---------|-------------| | pixel_size | 4 | Size of pixels in output (1-32). Higher = larger pixels | | upscale_after | true | Upscale output to match input resolution using nearest-neighbor | | copy_hue | false | Preserve original image hue (color tone) | | copy_sat | false | Preserve original image saturation (color intensity) | | copy_val | false | Preserve original image value (brightness) | | restore_dark | 15 | Blend dark areas with original (0-100%) | | restore_bright | 1 | Blend bright areas with original (0-100%) |

Tips

  • Best results: Use with regular photos or digital artwork, not existing pixel art
  • Color preservation: Enable copy_hue and copy_sat to keep original colors while applying pixel art style
  • Larger pixels: Increase pixel_size for a more retro look
  • Input size: Works best with images 256px+ in size

Limitations

  • Not designed for enhancing existing pixel art (may introduce artifacts)
  • Very small images may lack detail for good pixelization

Credits

  • Original Paper: Make Your Own Sprites (SIGGRAPH Asia)
  • Reference Implementation: https://github.com/arenatemp/pixelization_inference
  • Model Hosting: https://huggingface.co/ashleykleynhans/pixelization

License

This node wrapper is provided as-is. The underlying model is released under a non-commercial research license by the original authors.