SD3

SD3 (Stable Diffusion 3) is Stability AI's next-generation, open-source text-to-image model, utilizing a Diffusion Transformer architecture for superior image quality and text rendering.

SD3 represents a major leap in generative AI: Stability AI built it on a novel Diffusion Transformer (DiT) architecture, moving past the U-Net backbone. This model suite scales from 800 million to 8 billion parameters, offering flexibility for various compute needs and quality demands. Key performance gains include greatly improved adherence to complex, multi-subject prompts and a significant fix for a common industry issue: accurate in-image text and spelling. The technology leverages flow matching techniques for efficient training and generation, reinforcing Stability AI's commitment to democratizing access to high-fidelity, open-source models.

https://stability.ai/stable-diffusion-3