[19:02 Thu,22.February 2024 by Rudi Schmidts] |
Stability AI announced a few hours ago that a first beta version of Stable Diffusion 3 is now available. This is a powerful text-to-image AI model that is now said to have significant improvements in the areas of "performance on multiple topic queries", "image quality" and "spelling".
;) Stable Diffusion 3 - first beta presented The Stable Diffusion 3 suite comprises several models with a range of 800 million to 8 billion parameters (8B). With 8-bit quantization, even the largest model with a 12 GB graphics card should run without any problems. It is quite possible that even smaller graphics cards can provide sufficient memory for this with low quantization. ;) Stable Diffusion 3 - first beta presented In any case, Stability AI has the declared goal of making Stable Diffusion executable on as many computers as possible: "This approach reflects our core values and aims to democratize access to advanced AI." Users are given a variety of options in terms of scalability and quality to meet their creative needs." Stable Diffusion 3 combines a diffusion transformer architecture with flow matching. A detailed paper on this will be published shortly. If you look at the few  images on the Stability website published so far, the working font embedding in images stands out in particular. ;) Stable Diffusion 3 - first beta presented It usually doesn&t take long before far more results of a new model can be found on the web. We expect this to happen by the weekend at the latest...  more infos at bei stability.aideutsche Version dieser Seite: Stable Diffusion 3 - erste Beta vorgestellt |
 |
|
|