
“Top 15 Russian AI video generators” roundups consist almost entirely of services that give access to foreign models through a Russian-language interface. Russia has only a few video models of its own, and they share one trait: the clips are short and silent. Here’s what’s actually made in Russia, what it can do, and where to get models that generate sound.
Russian AI video generators: models and features
Data from Sber’s and Yandex’s help pages and the Kandinsky repository as of October 2, 2026.
| Kandinsky 5.0 (Sber) | Alice AI (Yandex) | Shedevrum (Yandex) | |
|---|---|---|---|
| What it does | text-to-video and image-to-video | animates photos | short clip from text |
| Where it works | GigaChat | Alice AI, Yandex Browser, alice.yandex.ru | Shedevrum app |
| Sound | no | no | not stated in the help page |
| Length and quality | 16:9 or 9:16 | 480p, up to 4 seconds | not stated in the help page |
| Free | yes, with a request limit | yes, with a limit; a boost costs 100 rubles | yes |
| Open weights | yes, MIT | no | no |
| Arena rank (Sept 21, 2026) | 37 and 44 of 48 | not ranked | not ranked |
Kandinsky 5.0: video in GigaChat and open weights
In GigaChat, video is generated by Kandinsky 5.0. Sber’s help page says you can generate clips for free right after signing up — from text or from an image, in 16:9 or 9:16. There’s a request limit, but the help page doesn’t give a number. The model doesn’t generate sound: Sber suggests adding music in a video editor. Sign-in is via Sber ID, only from a Russian IP and with VPN turned off.
What sets Kandinsky apart from every other model in this article is open weights. Sber published two video models on GitHub under the MIT license:
- Kandinsky 5.0 Video Lite — 2 billion parameters, fast, text-to-video and image-to-video;
- Kandinsky 5.0 Video Pro — 19 billion parameters, the best quality in the lineup.
According to the developers, in December 2025 Video Pro took first place among open text-to-video models on LMArena. Open weights let you run the model on your own server and build it into a product without paying for an API, but you’ll need a powerful GPU.
Yandex video: Alice AI and Shedevrum
Alice AI animates photos. In the Alice AI app, the Yandex app with Alice, Yandex Browser or at alice.yandex.ru, open “Studio” → “Animate,” upload a JPEG or PNG up to 10 MB and describe the motion. The help page states the specs plainly: 480p (854 × 480), up to 4 seconds, MP4, no sound. Once the free limit runs out, Yandex offers a “boost” for 100 rubles for extra generations per month.
Shedevrum makes short clips from a text prompt. You can only download the video as MP4 after publishing it in the Shedevrum feed — if the clip shouldn’t be public, this won’t work for you.
Yandex doesn’t disclose the name or version of its video model, and it isn’t on the Arena leaderboard. How to write prompts for animating photos in any model is covered in our guide animate a photo.
“Russian AI tools” in roundups are services, not models
Most entries in roundups are Russian aggregator services. Inside, they run foreign models: Veo, Kling, Seedance. What makes them “Russian” is a Russian-language interface, payment with Russian cards and access without a VPN. The clip’s quality is still determined by the model, not the service.
AI2Face is a service of this kind. Its generator includes video models from the top of the global leaderboard: Gemini Omni, Seedance, Veo 3.1 and Kling. They generate clips with sound — speech, ambient effects and music — from text and from photos. You can pay by card or crypto, credits from packs don’t expire, and you get welcome credits after signing up.
Russian AI models vs. the global leaders: the quality ranking
On Arena, users blindly compare two clips generated from the same prompt. As of September 21, 2026, the leaderboard lists 48 video models.
| Arena rank | Model | Developer | Sound | Where to use |
|---|---|---|---|---|
| 1–2 | Gemini Omni | yes | AI2Face | |
| 5 | Seedance 2.0 | ByteDance | yes | AI2Face |
| 12–24 | Veo (versions with audio) | yes | AI2Face | |
| 29 | Kling Pro | Kuaishou | yes | AI2Face |
| 37 and 44 | Kandinsky | Sber | no | GigaChat |
The gap isn’t just in the ranking. Russian models make silent clips, and Alice is also limited to 4 seconds at 480p. The leaders generate sound together with the picture and make clips of 8–30 seconds at 1080p and above. More about free video services is in our article free AI video generator, and about choosing a model for the task, in our overview of AI video generators.
Bottom line
Russian AI video generators are Kandinsky 5.0 by Sber and Yandex video in Alice AI and Shedevrum. They’re free, and Kandinsky can even be downloaded and run on your own hardware. They’re enough for short silent social media clips. If you need a clip with a voice, music and scene audio — for ads, YouTube or a client — there are no Russian options yet: use Gemini Omni, Seedance or Veo 3.1. In AI2Face, they’re all in one generator, paid by card or crypto.
FAQ
What Russian AI video generators are there?
Sber and Yandex have their own video models. Kandinsky 5.0 generates clips from text and photos in GigaChat, while Yandex animates photos in Alice AI and makes short clips in Shedevrum. Everything else in roundups is services that provide access to foreign models.
Is there a Russian AI video generator with sound?
Not among Russia's own models: Kandinsky 5.0 and Alice AI video don't generate audio. Veo 3.1, Seedance and Gemini Omni make clips with sound and speech — they're available in AI2Face.
Can I download a Russian AI video model?
Yes. Sber published Kandinsky 5.0 Video Lite (2 billion parameters) and Video Pro (19 billion) on GitHub under the MIT license. Running them requires a powerful GPU.
Which Russian AI video generator is the best?
Kandinsky is the only Russian video model on the Arena leaderboard: 37th and 44th out of 48. Among open models, Kandinsky 5.0 Video Pro ranked first in December 2025, according to its developers.
Sources
6
- https://alice.yandex.ru/support/ru/assistant/chat/picture-animate
- https://yandex.ru/support/shedevrum/ru/generate
- https://giga.chat/help/articles/kandinsky-to-create-a-video
- https://giga.chat/help/articles/faq
- https://github.com/kandinskylab/kandinsky-5
- https://arena.ai/leaderboard/text-to-video












