Stable Diffusion 3 Medium open-weight AI model released, NVIDIA and AMD optimizations

[harlan4096]

Stable Discussion 3 for image generation

The latest open-weight model is available for download. 



Stable Diffusion is one of the open-weight AI models for image generation. Unlike Midjourney or DALL-E, which are available through online APIs, this model can be downloaded and run locally using your graphics card. A more capable GPU with larger memory results in faster image generation and higher resolution outputs. However, this is not always the case, as optimizations and acceleration through advanced machine-learning libraries can significantly speed up the process.

Stability, AI’s Stable Diffusion 3 Medium is one of the most advanced (and at this point, most awaited) models for text-to-image generation. It is neither the smallest nor the largest model that Stability AI is working on, as this one has 2 billion parameters (there are also 800M and 8B versions planned); hence it is called Medium. Users can now download the SD3 model and use it for non-commercial purposes, while those without access can use Stability AI’s online tools.


Stable Diffusion 3 examples, Source: Stability AI

The key takeaways from the new SD3 model are its higher quality images through improved photorealism. It also has the ability to generate images based on more complex prompts and understand element compositions, spatial relationships, actions, and styles.

One of the key features of the new model is improved typography. Previously, this was an easy method to determine whether an image was generated by AI or created manually. This won’t be so easy anymore, as it can now generate text in various styles, and most of the time, it won’t make typos. Based on this example, it makes fewer typos than me (generated with StableSwarm and RTX 4090):
 
Stable Diffusion 3 Model, Source: VideoCardz
...

Continue Reading