Diffusers

Diffusers is a Python library for developers and researchers building image, video and audio generation into applications or running models on their own hardware. It supports pretrained generation pipelines as well as custom diffusion systems.

• Generation and editing: Generate images from text, edit images with prompts, fill selected regions, create image variations and upscale images using pipelines such as Stable Diffusion, ControlNet and InstructPix2Pix.
• Custom pipelines: Combine pretrained models with interchangeable noise schedulers, or compose pipeline blocks when you need control over the generation workflow.
• Training and adapters: Train and fine-tune diffusion models and adapters, and use LoRA adapters with pretrained pipelines.
• Memory and speed: Offloading and quantization help larger models fit on devices with limited memory; torch.compile can accelerate inference when memory allows.

Diffusers uses PyTorch and supports Apple Silicon. Pretrained checkpoints are available through the Hugging Face Hub, including Kandinsky and DeepFloyd IF models. The library is open source under Apache 2.0.