Toy Pixel-Art DDPM

A small diffusion project for learning how text-to-image models work. It generates 32ร—32 RGBA sprites from text prompts.

This is an educational toy model. Results can be noisy or miss the prompt, and image quality hasn't been thoroughly evaluated.

The U-Net was trained from scratch using a frozen CLIP text encoder. Training data: azyg0s/pixelart-172k.

Published weights: best validated EMA. Full settings are in config.json.

Try it

hf download Tobyz11/toy-pixelart-ddpm --local-dir toy-pixelart
cd toy-pixelart
pip install -r requirements.txt
python generate.py "pixel art icon of a purple dragon" --output sprite.png

Install huggingface_hub first if the hf command is missing. A GPU helps; CLIP downloads on the first run.

Downloads last month
38
Safetensors
Model size
24.5M params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Dataset used to train Tobyz11/toy-pixelart-ddpm