Just a few hours ago Stability AI he announced Stable Diffusion 3.
The new AI image generator follows the path already blazed by its predecessors, improving the processed content and offering a more specific understanding of users’ textual input.
The models of Stable Diffusion 3, according to its creators, can take into account from 800 million to 8 billion parameters to generate an image. This means visual outputs with an impressive amount of detail.
The brief announcement of Stability AI was not accompanied by any kind of demonstration, although the company opened a waiting list for the first and lucky ones tester of the new tool.
The CEO of the company described, at least in part, the potential of Stable Diffusion 3 Mother Mostaque. The same, through Xexplained how the tool uses a new system similar to what was proposed by Soracombined with several other improvements.
Stable Diffusion 3 CEO Emad Mostaque reveals more about the new tool
Stability, starting from 2022, has always taken great care of the models for generating AI images.
Starting from Stable Diffusion 1.4until XL Turbo and at the latest announcement, the company proved itself to be a more than worthy competitor to FROM-E 3 proposed by OpenAI. All this while also having to deal with various problems regarding copyright issues.
Returning to the latest version of this very powerful tool, Mostaque explained the new image creation system. This takes advantage of the architecture diffusion transform which, in fact, goes to work by combining them tiny pieces of photographs following some specific patterns and sequences (in addition to user instructions).
Furthermore, Stable Diffusion 3 adopts the technique known as flow matching, useful for generating complex images starting from confusing and approximate visual content. All this leads to impressive results from the point of view, with higher quality images than in the past.
Unfortunately, the absence of a real demo does not allow us to evaluate the real potential of the tool. The few generations published so far, at first glance, offer convincing but similar content to what was generated with the aforementioned FROM-E 3, Midjourney e Google Image. At first glance, Stable Diffusion 3 appears to handle prompts well.
According to Stability, once the tests are completed, the tool will be accessible to everyone for free. The company was keen to underline the great importance of this test phase, a period in which useful information will be collected to optimize performance and ensure safe use of the instrument.