LiliPixel V1 is a text-to-pixel-art model trained from scratch: a 191M-parameter diffusion transformer that draws sprites at a native 16x16 or 32x32 with a transparent background, or 64x64 scenes with an optional realistic mode. It needs about 1 GB of memory and runs with Diffusers or a ComfyUI node, with a free demo on Hugging Face. Trained on four public pixel-art datasets, two of them downscaled from FLUX images. Weights under OpenMDW-1.1. Complex scenes and text are weak.
Meet the team
Hunted by Ilyas. The makers are not on 25launches yet.