Meet a new level of creativity: Sber introduces an improved neural network for photo creation - Kandinsky 3.1

Comparing the capabilities of domestic artificial intelligence with foreign projects DALL-E 3 and Midjourney V6 is of interest to a wide audience. Sber's team announced the release of a new neural network model called Kandinsky. The updated version 3.1 will be available to all users in the near future, according to the bank's press service.

One of the key features of Kandinsky 3.1 was its training on an expanded database, which allowed the neural network to generate high quality images compared to the previous version 2.1. In addition, the speed of picture generation was increased by 10 times and the resolution of images was increased to 4K.

Other new features include the ability to generate text instead of abstract scribbles, the return of image style selection, and the ability to make localized changes without affecting the entire picture.

Kandinsky 3.1 has been given priority access to prominent artists and designers, which will likely help refine the neural network before its full launch. Sber also announced plans to unveil a new version of the AI for video generation called Kandinsky Video 1.1 within a year. According to the team, the resolution of the videos will be twice as high as the previous version of the program.