Stable Diffusion 3 Arrives - AI Art Gets Smarter, Safer

QUICK ZAPS

Stability AI has announced an early preview of Stable Diffusion 3, their latest and most capable text-to-image AI model. The new model shows improvements in its ability to understand multi-subject prompts and produce higher quality images with better spelling. 

While the full public release is still forthcoming, Stability AI is opening a waitlist for individuals to try out the early preview version. This preview phase allows them to gather user feedback that will help further enhance the model's performance and safety. The goal is to provide a range of options to meet different creative needs, with models ranging from 800M to 8B parameters in size.

Stable Diffusion 3 uses new diffusion transformer and flow matching architectures. Further details on its technical specifications will be released in an upcoming report. Stability AI emphasizes that safety is a top priority, from the model's training all the way through testing and deployment. Numerous safeguards have been implemented for the preview based on guidance from researchers and experts. 

By working with their community, Stability AI hopes to continue innovating generative AI tools with integrity. Their vision is to make AI openly and safely accessible to all. In the meantime, existing image generation and other AI services remain available through their membership and developer platforms.