In-Short
- Stability AI releases Stable Diffusion 3.5, enhancing AI image generation.
- Multiple model variants cater to different user needs and hardware capabilities.
- Stable Diffusion 3.5 Large features 8 billion parameters for high-resolution images.
- Models are free for non-commercial use and businesses with revenue under $1 million.
Summary of Stability AI’s New Release
Stability AI has introduced Stable Diffusion 3.5, a significant advancement in open-source AI image generation models. This release includes several variants to accommodate a range of users, from hobbyists to enterprise-level applications. The company has made improvements following feedback on the previous release, which did not fully meet community standards.
The Stable Diffusion 3.5 Large model is the most advanced, with 8 billion parameters and the ability to generate images at 1 megapixel resolution. A faster variant, Large Turbo, produces similar quality images but in fewer steps. A Medium version, suitable for consumer hardware, is also set to be released, featuring 2.5 billion parameters.
These models incorporate Query-Key Normalisation in transformer blocks to improve training stability and ease the fine-tuning process. However, this can lead to greater variation in outputs based on different seeds. Stability AI has also introduced a permissive community license, allowing free use for non-commercial purposes and for businesses with annual revenues below $1 million.
Access to the latest image generation models is available through platforms such as Hugging Face, GitHub, the Stability AI API, Replicate, ComfyUI, and DeepInfra. For more information and to explore the capabilities of Stable Diffusion 3.5, interested parties can visit the official sources.
Image Credit: Stability AI
For a deeper dive into Stability AI’s Stable Diffusion 3.5 and its capabilities, please visit the original source.
