This repository implements Flash Diffusion, a method for accelerating any conditional diffusion model through distillation to enable high-quality image generation in very few sampling steps. The project provides pre-trained checkpoints for Stable Diffusion, SDXL, PixArt, and SD3 variants, supporting tasks including text-to-image, inpainting, and super-resolution. It is published as an AAAI 2025 Oral paper with official Hugging Face model weights and demo spaces.