Synthetic media startup Stability AI has launched SDXL 0.9, which is an enterprise-focused upgrade of its generative AI model, Stable Diffusion. The model is designed to generate realistic images from text prompts with better composition and improved photo-realism. SDXL 0.9 has one of the largest parameter counts of any open-source image model, with a 3.5 billion parameter base model and a 6.6 billion parameter model ensemble pipeline. The upgrade expands creative use cases for generative AI imagery and offers advancements for design and industrial use, Stability AI noted.
The upgrade allows the AI model to add finer details to the generated output of the first stage, including the correct spelling of words, and the fine details of faces and hands that other generative AI image models often struggle to depict. This is the result of the model’s increased parameter size since the beta, with the initial test version having 3.1 billion parameters and relying on a single model instead of two CLIP models as in the upgraded version. SDXL 0.9 is available on ClipDrop, the app Stability AI acquired last year, with an API in development to embed the tool in existing software.
Stability AI has been rapidly expanding its portfolio since raising $101m in October, with the recent addition of cartoon-making Stable Animation SDK and the non-visual generative AI tool, StableLM, which is capable of composing text and computer code. The release of SDXL 0.9 and other models demonstrates the company’s commitment to developing AI technologies that can cater to different enterprise needs.