Why it matters: Stability AI has released its most advanced image generation models to date, promising improved diversity in outputs and better prompt adherence. This release marks a significant step forward in making AI image generation more inclusive and accessible while addressing previous criticisms about representation and quality.
The Innovation: The new Stable Diffusion 3.5 series introduces three distinct models, each targeting different use cases. The flagship Stable Diffusion 3.5 Large boasts 8 billion parameters and can generate professional-quality images at 1-megapixel resolution.
- Large Turbo variant offers faster generation in just four steps (Techcrunch)
- Medium version optimized for consumer devices coming October 29 (Engadget)
Technical Advances: The models incorporate new features to improve image quality and diversity, including Query-Key Normalization for better training stability. Stability AI claims the models can generate more representative outputs without requiring specific prompting.
- Enhanced prompt adherence compared to previous versions
- Improved handling of different visual styles and aesthetics
Accessibility: Following recent controversies over AI image generation and representation, Stability AI has made these models available under a permissive community license. The models are free for non-commercial use and for businesses earning less than $1 million annually.