Sber opens Kandinsky Video to the public

Sber has released its Kandinsky Video model to the public through the GigaChat and Kandinsky bot on Telegram. The feature is in the bot's "Create video" menu.

Under the access terms in the company's blog post, anyone can generate 10 videos for free as a trial. Users who sign in can then generate 10 videos a day. Sber is also running a referral scheme: each person who joins through a user's invite link adds 10 generations to that user's daily limit, so a user who brings in two friends can make up to 30 videos a day.

The model supports text-to-video and image-to-video. In image-to-video mode, it takes the scene from an uploaded picture, so a short motion prompt such as "a cat yawns" is often enough. Sber's examples include animating children's drawings, old family photos, book illustrations and a marble bust. Users can also choose cinematic camera moves, including zooming in or out and rotating around a subject.

The post also suggests chaining Sber's own tools. Users can create stills in Kandinsky Image, animate them in Kandinsky Video and cut the clips into a longer video. Sber's sample Arctic montage uses music from SymFormer, another Sber model.

Most of the announcement is prompting advice and sample outputs. It does not give the model's architecture, size, resolution, clip length or benchmark results, and it does not compare Kandinsky Video with other video generators.