huggingface/diffusers

Diffusers is a Python library from Hugging Face that provides ready-to-use implementations of diffusion models, the AI technology behind tools like Stable Diffusion that generate images, videos, and audio from text descriptions. A diffusion model works by learning to gradually remove noise from a random signal, starting with pure static and iteratively refining it into a coherent image, audio clip, or video frame guided by a text prompt or other input. The library is built around three modular building blocks. Pipelines are high-level objects that combine everything needed for a specific task (such as text-to-image generation) into a single easy-to-use interface, you can generate an image with just a few lines of code by loading a pretrained model from Hugging Face's model hub. Schedulers control the noise-removal process at inference time, trading speed against quality. Models are the neural network components (like UNet architectures) that can be combined in custom ways to build specialized pipelines from scratch. Someone would use Diffusers when they want to run or experiment with AI image generation locally, fine-tune a pretrained model on their own images, or build a custom image generation application. It supports both simple inference use cases (loading a model and generating images) and advanced research workflows (training new models or modifying architectures). The tech stack is Python with PyTorch as the deep learning framework. It also supports Apple Silicon (M1/M2) via the MPS backend and works with CUDA GPUs. Models from over 30,000 checkpoints on the Hugging Face Hub can be loaded directly.