Open source company Stability AI launches Stable Diffusion 3, a new next-generation text-to-image generator with improved performance. The preview version is available via a waitlist.

Stability AI launches Stable Diffusion 3, a next-generation text-to-image generator with improved performance that can process multiple objects in a single prompt. This model is the next in line of AI models from the open source company (the last version was Stable Diffusion XL Turbo), which also brought Stable Audio to the AI ​​world last year. The model is not yet publicly available, but Stability AI has opened a waiting list for users who want to try it out. The model is based on two techniques: diffusion transformer architecture and flux matching. Based on insights from the preview phase, Stability AI can improve performance and security.

Open source

Stability AI is a company that follows the open source principle, which is also anchored in the new Stable Diffusion 3. This model range currently ranges from 800M to 8B parameters. The AI ​​company wants to offer users a variety of scalability and quality options to meet creative needs. The company is committed to making generative AI safely accessible to everyone.

The Stable Diffusion 3 Series utilizes the diffusion transformer architecture combined with flow matching. The former approach is based on transformers that can handle patterns and sequences well. In this way, the AI ​​swaps the usual faceplates for a system that edits small parts of the image. Flow matching is a technique that allows AI models to generate images by learning how to smoothly transition from noise to a structured image. This new approach is intended to provide higher quality images.

The company is committed to making generative AI safely accessible to everyone. This open source nature has led to a lawsuit against Stability AI in the past for using copyrighted material without permission.