All Tools
P
OtherFreeOpen Source
PIXART
Fast DiT training for photorealistic text-to-image
Apache-2.0
ABOUT
Large UNet diffusion models are slow to train and heavy to sample. PixArt-alpha trains a 0.6B DiT on T5 captions so labs can reach photorealistic 1024 images, LCM few-step sampling, and ControlNet without a Stable Diffusion-scale compute budget.
INTEGRATION GUIDE
1. Sample PixArt-XL-2-1024 with Diffusers PixArtAlphaPipeline
2. Train LoRA or LCM adapters on custom captions with the released scripts
3. Run PixArt ControlNet for layout-conditioned generation
4. Generate images in ComfyUI from the ExtraModels PixArt nodes
TAGS
pythondiffusiontext-to-imagetransformerdiffuserscomfyuiiclr