YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
license: cc-by-nc-4.0 library_name: ai2pixelart pipeline_tag: image-to-image tags: - pixel-art - image-restoration - image-to-image
ai2pixelart — model weights
Trained restoration networks for ai2pixelart — turning AI-generated pseudo pixel art (wobbly grids, mixed-color cells, too many shades) into pixel-perfect, palette-clean pixel art.
The networks assign each cell a palette entry from the classical grid/palette proposal; they are structurally incapable of off-palette colors and carry learned priors (recovering pure 1-px details, collapsing shaded backgrounds) that no local method can.
Models
| File | Approach | Use |
|---|---|---|
robust-v11.safetensors |
Neural Robust | Recommended default; best on real AI renders, dense palettes, noisy backgrounds, tile sheets. |
detail-v9.safetensors |
Neural Detail | Best 1-px detail retention on fine 2–3 px grids; fallback when Robust over-corrects. |
Each file is a self-describing safetensors checkpoint (model config in the metadata header).
Usage
pip install "ai2pixelart[neural]"
ai2pixelart clean input.png -o out.png --approach robust # or: detail
The weights download automatically on first use and are cached locally.
License
Licensed under CC BY-NC 4.0, with an important carve-out: the pixel
art you make with these models is yours — you may use it commercially.
The noncommercial term restricts the models themselves (you may not sell
them or offer their functionality as a paid product/service/API), not your
output. See LICENSE-WEIGHTS in this repo.
The source code is licensed separately under the PolyForm Noncommercial License 1.0.0 (see the code repository).