On October 6, 2026, Sber rolled out video generation with sound in GigaChat. The AI now creates both visuals and audio—including speech, music, and sound effects—in a single process. This feature is free for GigaChat users, with no credit card or workarounds required. The resulting clips are up to five seconds long and can be rendered in up to Full HD quality.
For short-form video creators, this marks the arrival of a new tool, accessible from Russia, for creating B-roll, intros, and draft UGC scenes. Let's break down what it can and can't do, and how to integrate these five-second clips into a full video.
What's New
The generation is powered by Sber's new model, Kandinsky 6.0 Video. Its key difference from previous versions is that the visuals and audio are created together by a single model, not stitched together afterward. This means a character's lips sync with their lines, footsteps match their stride, and the music changes with the scene.
A scene can be prompted in two ways:
- Describe it with a text prompt—what's happening, who is speaking, and what sounds are needed.
- Upload a starting frame (a photo or image) and separately describe the desired action and sound.
Audio can also be disabled, as the model is capable of generating silent videos.
Specs: What Sber Promises
| Parameter | Value |
|---|---|
| Video length with sound | up to 5 seconds |
| Quality | SD, HD, or Full HD (up to 1920×1080) |
| Audio sample rate | 44 kHz |
| Frame rate | approx. 24 fps |
| Audio features | lip-synced speech, music, ambient sounds, effects |
| Price for GigaChat users | free |
| Generation time per video | approx. 8 minutes (according to ppc.world) |
Five seconds is the current limit. Sber says it plans to increase it but has not given a timeline.
How It Stacks Up Against the Competition
It's important to note the source of these figures. All comparisons below are Sber's own claims; there are no independent benchmarks yet:
- The new model outperforms the previous version, Kandinsky 5.0, in about 71% of cases across all criteria.
- In terms of visual quality and motion accuracy, it surpasses the open-source model LTX 2.5 and the closed-source Veo 3.1 Fast.
- The team's research paper is more cautious, stating that Kandinsky 6.0 Video outperforms LTX 2.5 on most VABench test metrics and "remains competitive" with closed-source models, especially in speech quality.
Regarding training data, Sber reported using "over 20 million videos with sound." The team's paper provides a more detailed breakdown: 50 million images, 20 million video scenes, 40 million audio tracks, and 7 million segments with synchronized video and audio.
The Model is Also Free for Developers
In a related announcement, Sber has also open-sourced the model under the MIT license. This is one of the most permissive common licenses, allowing for use in commercial products.
"We are releasing the Kandinsky 6.0 Video model to developers for free and without restrictions," said Anton Frolov, Senior Vice President and Head of Generative AI Development at Sberbank.
What's been released:
| What | Details |
|---|---|
| Model versions | Lite — 3 billion parameters, Pro — approx. 30 billion |
| Upscaling to Full HD | A separate 1-billion-parameter model for ×2 and ×4 upscaling |
| Where to find it | Hugging Face and GitHub |
| Integrations | Diffusers, FAL, FastVideo, ComfyUI |
| Local setup requirements | 16 GB of VRAM or more (according to Habr); the model card mentions offloading parts to RAM to fit on smaller cards |
A useful technical detail: the model itself generates video at a lower resolution—864×480. The Full HD quality comes from a separate upscaling model. This means 'Full HD' here is an enlarged image, not a native 1080p render. The difference is barely noticeable in short inserts, but if you stretch a frame to full screen, the softness becomes apparent.
How Short-Form Video Creators Can Use It
Five seconds isn't a full video; it's a component. Here's where these clips can really shine:
Intros and Outros
A five-second scene with the right mood and sound makes for a solid intro for a video series. Create it once and use it in every episode.
B-Roll for Voiceovers
When you're narrating something but have nothing to show, a five-second scene can fill the gap. In this case, it's best to mute the model's audio and keep your own voiceover.
Drafts for Clients
If you're creating UGC and need to get an idea approved, it's easier to show a five-second scene than to explain it in words. The actual shoot goes faster because everyone has already seen what you meant.
Animating Photos
The "start frame + description" mode is a great way to bring a static image to life. For archival photos, screenshots, and product shots, this can be more effective than generating from scratch.
The editing process is simple: generate 4–6 different five-second scenes, assemble them into a single video, and lay a consistent audio track over the top. This way, the viewer doesn't notice the length of each individual clip, and the music helps mask any stylistic differences between the scenes.
A Note on Disclosure: Don't Forget to Label
TikTok, YouTube, and Instagram require creators to label realistic AI-generated content. This label is applied on the platform itself during upload, not within GigaChat. If you create an insert that could be mistaken for real footage, it's best to add the disclosure: platforms may reduce the reach of unlabeled content and, in serious cases, remove monetization. Abstract graphics and clearly illustrated scenes do not fall under this requirement.
What We Still Don't Know
Let's be upfront about what Sber hasn't clarified:
- Daily generation limits for free access. The press release says "free" but doesn't specify a number. We'll have to find out through trial and error.
- Future paid plans. Nothing has been mentioned.
- Vertical format. The model cards and examples show horizontal resolutions (864×480 and 1920×1080). There has been no announcement about the 9:16 aspect ratio needed for Shorts, Reels, and Clips. For now, assume you'll have to crop the horizontal frame.
- Terms of use for videos from GigaChat. The MIT license applies to the downloadable model itself. What you can do with videos created within the GigaChat service is governed by the service's terms—no specific clarification was issued with this launch.
- When the 5-second limit will be increased. Plans have been announced, but no timeline has been given.
If any of these points are critical for your workflow, don't build your plans around them until Sber makes a direct announcement.
FAQ
What free AI video generator works in Russia?
GigaChat with the Kandinsky 6.0 Video model. It's a Russian service that can be accessed without workarounds, and video generation is free. We covered alternatives and their availability in our article Sora is Shut Down: Why and What to Use Instead — Alternatives for AI Video in Shorts, Reels, and UGC.
How do I make a video with sound in GigaChat?
Describe the scene in a text prompt and specify what should be heard: who is speaking and what they're saying, the type of music, and any sound effects. Alternatively, upload a starting frame and add a description of the action and audio. Choose your quality—SD, HD, or Full HD. A single generation takes about eight minutes.
How long are videos from GigaChat?
Up to five seconds with sound. An increase to this limit has been promised, but no date has been given. To create a full video, you'll need to edit multiple five-second scenes together.
Can I use these videos on TikTok, Shorts, and Reels?
Technically, yes. There are two things to keep in mind: the frame is horizontal, so you'll need to crop it for vertical formats, and you must label any realistic AI scenes when uploading to the platform. We've compiled a list of tools for assembling and polishing your videos in our article AI for Video Editing: 15 Best AI Editors of 2026, Free and Paid.
Where can I download Kandinsky 6.0 Video and what do I need to run it?
The weights and code are available on Hugging Face and GitHub under the MIT license, with integrations for Diffusers, ComfyUI, FAL, and FastVideo. According to Habr, you'll need at least 16 GB of VRAM for local setup; the 3-billion-parameter Lite version requires less than the Pro version. If you don't have a powerful enough GPU, it's easier to use GigaChat.
Do I need subtitles if the AI generates the audio?
Yes. The AI-generated speech is short, and short-form videos are often watched without sound. We wrote about how to create subtitles for free here: Free Video Subtitles: X Now Edits and Translates Them with Grok.
Free tools like this lower the barrier to entry: a scene that once required a paid shoot can now be generated in about eight minutes. On Prime Oracles, briefs from brands are posted openly. In the "Content Rewards" section, you can view the terms and CPV rates, and accept a brief by clicking the "Accept Brief" button.
Sources: Sber press release (CNews), ixbt, Habr, ppc.world, models on Hugging Face, team's research paper.