Stability AI Launches its "Most Capable" Text-to-Image Model Stable Diffusion 3

British AI startup Stability AI has introduced Stable Diffusion 3, its latest text-to-image model currently in early preview. This model boasts significant enhancements in handling multi-subject prompts, improving image quality, and refining spelling abilities. 

Although not widely accessible yet, the platform is now accepting sign-ups for the waitlist to participate in the early preview phase. This stage is vital for collecting insights to enhance performance and ensure safety before a broader release. Interested users can register for the waitlist to gain access.

What does the Stable Diffusion 3 model consist of?

The Stable Diffusion 3 suite comprises models ranging from 800M to 8B parameters, aligning with its commitment to inclusivity and accessibility. This diversity allows users to choose options based on scalability and quality, catering to various creative requirements. The model's architecture combines diffusion transformer and flow matching, details of which will be elaborated in an upcoming technical report.

Focus on Responsible AI

“We believe in safe, responsible AI practices,” read a company blog post.

The company claims that its dedication to safe and responsible AI practices is evident throughout the development lifecycle of Stable Diffusion 3. From initial training to testing, evaluation, and deployment, the researchers implement rigorous measures to prevent misuse by malicious actors.

In anticipation of the early preview, it has incorporated multiple safeguards.

“Our commitment to ensuring generative AI is open, safe, and universally accessible remains steadfast. With Stable Diffusion 3, we strive to offer adaptable solutions that enable individuals, developers, and enterprises to unleash their creativity, aligning with our mission to activate humanity’s potential,” the blog concludes.