Users of the GigaChat service can now generate video clips that include speech, music, and ambient sound effects. This functionality is powered by the newly integrated Kandinsky 6.0 Video model developed by Sber.
The system allows users to create videos by providing a text description or by uploading an initial frame. Users can specify the desired audio accompaniment, or choose to generate a video without any sound at all.
Kandinsky 6.0 Video supports resolutions up to Full HD, with options for SD and HD quality depending on speed requirements. Currently, the maximum duration for a video clip with synchronized audio is five seconds.
The model features improved synchronization between visual elements and audio, including lip-syncing for characters and matching ambient noises like footsteps, wind, or traffic to the on-screen action.
Sber reports enhancements in the rendering of physics and object movement, resulting in more natural depictions of people, animals, and machinery during dynamic sequences.
The model is available to developers under the MIT open-source license, allowing for integration into third-party products via standard generative AI development tools.
