huggingface/Diffusers
🤗 Diffusers: State-of-the-art diffusion models for image, video, and audio generation in PyTorch.
Diffusers is the open-source Python library from Hugging Face for running and training state-of-the-art diffusion models that generate images, video, and audio. It turns models like Stable Diffusion, FLUX, and Qwen-Image into a few lines of code, backed by 30,000+ checkpoints on the Hub.
Our Review
Diffusers is the most widely used library for diffusion-based generation, with more than 33,000 GitHub stars and millions of monthly PyPI installs. Hugging Face maintains it as the reference implementation that most new image and video models ship with on day one.
What Diffusers does:
- •One-line model loading
DiffusionPipeline.from_pretrained(...)downloads and runs any of 30,000+ checkpoints from the Hugging Face Hub. - •Image, video, and audio generation text-to-image, image-to-image, text-to-video, image-to-video, inpainting, and audio pipelines in one toolkit.
- •Latest model support Stable Diffusion, FLUX, and Qwen-Image are added on or near their release dates.
- •Interchangeable schedulers swap noise schedulers to trade generation speed against output quality without changing the model.
- •Composable building blocks use pretrained models and schedulers directly to build custom end-to-end diffusion systems.
- •Training and fine-tuning the same library supports training your own models, not only running inference.
Diffusers ecosystem:
- •Hugging Face Hub 30,000+ diffusion checkpoints browsable and loadable by name.
- •Transformers and Accelerate pairs with the rest of the Hugging Face stack for multi-GPU and optimized inference.
- •PyTorch native built on PyTorch, with documented Apple Silicon (M-series) support.
Getting started:
Install with pip install --upgrade diffusers[torch] (or conda-forge). Load a model with DiffusionPipeline.from_pretrained(...), move it to your GPU, and call it with a prompt to get an image. Full docs and a model browser live at huggingface.co/docs/diffusers. Release v0.38.0 (May 2026) added new image and audio pipelines.
Limitations:
Diffusers is a developer library, not a point-and-click app, so using it means writing Python and having a capable GPU for fast generation (large models want significant VRAM). Hugging Face documents a deliberate bias toward usability over raw performance, so a hand-tuned inference server can be faster. With 1,000+ open issues and a fast release cadence, the newest pipelines can have rough edges. There is no built-in interface; tools like Open WebUI or ComfyUI wrap it for non-coders.
Our Verdict
Diffusers is the default way to run open diffusion models in code, and in 2026 it remains the library that new image and video models target first.
For Developers, it is the fastest path from a model name to a generated image: one from_pretrained call, a prompt, and a GPU. The modular pipeline, scheduler, and model design lets you start with a turnkey pipeline and drop down to building blocks when you need control over the generation loop.
For Vibe Builders, Diffusers is the engine under many of the image tools you already use, but calling it directly means writing Python and managing a GPU. If you want results without code, a wrapper like Open WebUI or a hosted API is a better fit, while Diffusers stays the right call once you need programmatic control or custom training.
Skip it if you only want a no-code image generator or you have no access to a GPU, because Diffusers is a developer library rather than an application. For production image serving at scale, pair it with optimized runtimes rather than the default pipelines.
Frequently Asked Questions
Is Hugging Face Diffusers free?
Yes, Diffusers is free and open-source under the Apache 2.0 license, and Hugging Face maintains it on GitHub where it has more than 33,000 stars as of 2026. You can use it commercially at no cost, and you only pay for the compute, whether that is your own GPU or a cloud GPU, that runs the models.
What can you generate with Diffusers?
Diffusers supports text-to-image, image-to-image, inpainting, text-to-video, image-to-video, and audio generation. It runs state-of-the-art models including Stable Diffusion, FLUX, and Qwen-Image, and it gives access to more than 30,000 pretrained checkpoints on the Hugging Face Hub. The same library also supports training and fine-tuning your own models.
Do I need a GPU to use Diffusers?
A GPU is strongly recommended because diffusion models are compute-heavy and run slowly on a CPU. For fast image generation you want an NVIDIA GPU with enough VRAM for your chosen model, though Diffusers also documents Apple Silicon support. As an alternative, you can run it on a rented cloud GPU instance.
What is the difference between Diffusers and ComfyUI?
Diffusers is a Python library you call from code, while ComfyUI is a node-based graphical interface for building generation workflows. Choose Diffusers when you want programmatic control, training, or to embed generation inside an app. Choose ComfyUI when you prefer a visual, no-code way to wire up pipelines.
What models does Diffusers support?
Diffusers supports a broad and growing set of diffusion architectures, including Stable Diffusion, FLUX, Qwen-Image, and many latent-diffusion and video models. New architectures are typically added on or near their release. In total you can load more than 30,000 community and official checkpoints from the Hugging Face Hub.
What is Diffusers?
Diffusers is the open-source Python library from Hugging Face for running and training state-of-the-art diffusion models that generate images, video, and audio. It turns models like Stable Diffusion, FLUX, and Qwen-Image into a few lines of code, backed by 30,000+ checkpoints on the Hub.
How do I install Diffusers?
Visit the GitHub repository at https://github.com/huggingface/diffusers for installation instructions.
What license does Diffusers use?
Diffusers uses the Apache-2.0 license.
What are alternatives to Diffusers?
Explore related tools and alternatives on My AI Guide.
Open source & community-verified
Apache-2.0 licensed: free to use in any project, no strings attached. 33,792 developers have starred this, meaning the community has reviewed and trusted it.
Reviewed by My AI Guide for relevance, quality, and active maintenance before listing.
Topics