Kandinsky 6.0 Video: Sber's First Neural Network with Sound

Sber has unveiled the Kandinsky 6.0 Video model, capable of generating video clips with sound immediately after creation. All that is required is a text description of the scene or an uploaded first frame.

The neural network generates Full HD video with synchronized audio: speech, music, and environmental sounds. Current clips are limited to five seconds, but developers plan to extend their duration.

The model is already available for developers in open access. According to Anton Frolov, head of Sber's Generative AI block, artificial intelligence should become as natural a tool as a camera or a text editor.