We have reached a point where our eyes are not enough to discern what is true and what is not. While governments and administrations propose solutions, AI continues its path without looking back.
.png){kind=icon}
On Thursday, Stability AI announced Stable Diffusion 3, a next-generation open-weight image synthesis model. Building on its predecessors, it generates detailed images of multiple subjects with higher quality and accuracy in text generation.
The brief and unclear announcement has not been accompanied by a public demonstration, but Stability opens today a waiting list for those who wish to try it.
What’s new in Stable Diffusion 3
Stability claims that its family of models Stable Diffusion 3 (which takes text descriptions called “prompts” and converts them into matching images) has a size of between 800 million and 8 billion parameters.
This range of sizes allows different versions of the model to run locally on different devices, from smartphones to servers. The size of the parameters roughly corresponds to the model’s capacity in terms of the amount of detail it can generate.
Since 2022, we have seen Stability release a progression of AI image generation models: Stable Diffusion 1.4, 1.5, 2.0, 2.1, XL, XL Turbo, and now Stable Diffusion 3.
Why does Stable Diffusion work when DALL-E exists
Stability has made a name for itself by offering a more open alternative to proprietary image synthesis models, such as OpenAI’s DALL-E 3, although not without controversy due to the use of copyrighted training data, bias, and potential for abuse.
Stable Diffusion models are open-weighted and open-source, which means that the models can be run locally and adjusted to change their results.
Regarding technological improvements, Stability’s CEO, Emad Mostaque, wrote on Twitter: “This uses a new type of broadcast transformer (similar to Sora) combined with flow adjustment and other enhancements. This leverages the transformer improvements and can not only scale further, but also accept multimodal inputs.”
As Mostaque said, the Stable Diffusion 3 family uses the diffusion transformer architecture, which is a new way of creating AI-generated images that replaces the usual building blocks of images with a system that works on small fragments of the image.
The method is inspired by transformers, which are good at handling patterns and sequences. This method is not only effective at scale, but also produces higher quality images.
