Stability AI has launched a first preview of its new text-to-image model, Stable Diffusion 3. With this preview, the AI developer primarily aims to gather feedback.
In the preview, Stability AI introduces the latest iteration of its text-to-image model. According to the AI developer, this version brings “even more potential” for creatives to express their creativity.
Stable Diffusion 3 offers new capabilities for improved performance when generating images with prompts. Among other things, this version supports multi-subject prompts and improves both the quality of generated images and backlash.
The various AI models that form the basis of Stable Diffusion 3 contain between 800 million and 8 billion parameters. According to Stability AI, this size is aligned with standards and available to anyone following the AI developer. This allows users to choose different options for scalability and quality based on their needs.
In addition, the Stable Diffusion 3 model uses a diffusion transformer architecture and flow matching. The AI developer promises to make more technical details available soon.
The new version of the text-to-image model also includes security settings aimed at responsible AI development. However, Stability AI has not disclosed the specific security settings.
Waitlist
Users interested in testing Stable Diffusion 3 can sign up for a waiting list. The AI developer is releasing the preview to gather as much feedback as possible, which can be used to improve the text-to-image model in terms of both functionality and security before the final version is released.
Also read: Stability AI seeks possible buyer due to dispute with shareholder