Diffusers

Diffusers is Hugging Face's comprehensive library for state-of-the-art pretrained diffusion models for image, video, and audio generation in PyTorch. Supporting cutting-edge models like FLUX (12B parameters), Stable Diffusion, and dozens of other architectures, Diffusers provides a unified interface for generative AI tasks. The 2025 version features massively refactored attention mechanisms supporting Flash Attention 3, SAGE attention, and PyTorch's scaled_dot_product_attention, along with single-file modeling implementations following Transformers' philosophy. Optimized for memory-constrained hardware and accelerated inference on GPUs, CPUs, and TPUs, Diffusers enables everything from text-to-image generation to advanced image editing with multi-step consistency.