Stable Diffusion XL

Stable Diffusion XL is an image generation model for creators and developers who want to produce images on their own hardware or servers. You can start with a text prompt or use an existing image as input.

• Text-to-image generation: Create images from written prompts with the SDXL base model.
• Image-to-image generation: Use an existing image as the starting point for a generated image.
• Generation interface: The supplied Streamlit demo gives you access to both generation modes, with support for SDXL-base and SDXL-refiner.
• Model development: Developers can use training examples and modular components to build or adapt diffusion models in PyTorch.

The Python code runs in your own environment and is open source under the MIT License. Model weights carry separate terms: SDXL 1.0 weights use CreativeML Open RAIL++-M, while SDXL 0.9 weights use a research license. The MIT code license doesn't replace either set of weight terms.

Stability AI also offers managed applications and media generation APIs. Those services run outside your machine and are separate from self-hosting the model.

Images generated through the supplied code contain an invisible watermark. A detection script can check for that watermark.

The downloadable SDXL Base 1.0 and Refiner 1.0 checkpoints are fixed model releases. They use CreativeML OpenRAIL++-M, while the shared generative-models reference code uses MIT. A release of that shared codebase is not the SDXL model version.