Sber Makes Kandinsky Avatar 1.0 Available to Developers
Sber has made its Kandinsky Avatar 1.0 family of Russian AI models for creating talking video avatars publicly available. Developers can use the models in commercial services.
The model generates videos of up to one minute at 24 frames per second and can animate people, animals, and fictional characters. Its code and weights are available free of charge under the MIT License. To create an avatar, users need only an image, an audio recording, and a scene description. The system synchronizes lip movements with speech and adds facial expressions and gestures. It does not clone voices. Kandinsky Avatar is built on the Pro and Lite versions of Kandinsky 5.0 Video.
According to Dimitrov, developers will be able to integrate the model into their own products and fine-tune it for specific tasks without having to build the technology from scratch. Feedback from the developer community will help identify how the model performs across different use cases and what needs to be improved. Sber plans to use these insights in future versions of Kandinsky and when integrating video avatars into its services.
