Stability AI releases Stable Diffusion XL

Stability AI released Stable Diffusion XL 1.0, an image-generation model combining a 3.5-billion-parameter base model with a 6.6-billion-parameter refinement stage in a two-part pipeline, and made it available as open weights alongside hosted access through its Clipdrop and DreamStudio products and API.

Stability said SDXL improved over prior Stable Diffusion releases on several fronts that had been persistent weaknesses of diffusion image models: more accurate colour and contrast, better handling of difficult subjects such as hands and rendered text, native generation at 1024×1024 resolution rather than 512×512, and prompts that no longer needed padding with quality keywords such as “masterpiece” to produce polished results. The company cited internal preference testing in which evaluators favoured SDXL’s outputs over those of other open image models available at the time, though as with most such comparisons this was a company-run study rather than an independently verified benchmark.

The model was released under the CreativeML OpenRAIL++-M licence, a permissive-but-restricted licence that allowed commercial use while prohibiting a defined set of harmful applications — the same licensing approach Stability had used for earlier Stable Diffusion releases, intended to balance broad availability against liability for misuse.

SDXL’s release consolidated Stability’s position as the main open-weight counterpart to closed, API-only image generators such as Midjourney and DALL-E: because the weights were downloadable rather than accessed only through a hosted service, developers could fine-tune, host and modify the model themselves, and an ecosystem of community fine-tunes and tools built on SDXL specifically grew over the following year, extending its practical reach well beyond Stability’s own products.