Stable Diffusion 3 is Released
Stability AI has just unveiled an early preview of Stable Diffusion 3. This new release is touted as the most capable text-to-image model yet. It shows significant improvements in handling multi-subject prompts, delivering higher image quality, and enhanced spelling abilities.
Currently, Stable Diffusion 3 is not widely available to the public. However, interested users can join a waitlist starting today. The purpose of this limited release is to collect valuable insights that will aid in refining the model’s performance and safety before it becomes openly accessible.
Stable Diffusion 3 encompasses a range of model sizes, starting from 800 million to an impressive 8 billion parameters. Its architecture uniquely combines diffusion transformer architecture with flow matching. Stability AI has also stated that a detailed technical report on Stable Diffusion 3 will be published soon, providing further insights into its advancements.
Stability AI is firmly committed to preventing misuse by bad actors. This commitment spans across all stages: training, testing, evaluation, and deployment of the model. Numerous safeguards have been introduced specifically for this early preview phase. Moreover, there’s an ongoing collaboration with researchers, experts, and the broader community aimed at ensuring innovation proceeds with integrity.
Stability AI further reaffirms its dedication to principles of openness, safety, and universal accessibility in generative AI technologies. Aligned with its mission “to activate humanity’s potential,” the company offers adaptable solutions designed to inspire creativity across various sectors.
For commercial purposes, individuals and organizations are encouraged to visit the Stability AI Membership page or explore its Developer Platform for more tailored options.
This release marks yet another major milestone for Stability AI as it continues to lead in innovating accessible AI tools for creative endeavors worldwide.
