Choose Kandinsky 6.0 Video when…
Choose Kandinsky 6.0 Video if you want weights you control: running it on your own NVIDIA GPUs, fine-tuning it, or building it into a product without per-clip fees, and 5-second clips are enough.
Video model alternative
Kandinsky 6.0 Video Alternative: Open Weights vs a Hosted Model on Imgo
Kandinsky 6.0 Video is Sber's open-weight video model with synchronized audio, released by Kandinsky Lab on October 6, 2026 under the MIT license. It is not available on Imgo. Compare what running it yourself involves with Seedance 2.0, which you can run on Imgo today.
External reference — not available on ImgoLast checked: 2026-10-23
Reviewed against linked first-party pages. Plan limits and model availability can change.
Compare the working requirements
Choose Imgo when…
Choose Imgo when you need longer clips without running GPUs: Seedance 2.0 in Imgo's video generator renders 5 to 15 seconds up to 4K from text, a start frame, a start and end frame, or up to 4 reference images, with native audio. Kandinsky 6.0 Video itself is not available on Imgo.
Compare the working requirements
The table describes documented capabilities and checks to make, without assigning unsupported scores or assuming a plan includes every tool.
| Decision factor | What to verify |
|---|---|
| Best suited to | Creators and developers weighing an open-source audio-video model they host themselves against a hosted model they can use in the browser now. |
| Image and video | Kandinsky 6.0 Video is a video-with-audio model you run through Diffusers, ComfyUI, vLLM-Omni, SGLang or FastVideo, or try in Sber's GigaChat. Imgo hosts several video models, plus image generation and editing for the frames that feed them. |
| References and editing | Kandinsky 6.0 Video takes a text prompt, or one image plus a prompt; the repository lists no end-frame, reference-video or reference-audio input. Seedance 2.0 on Imgo takes a start and end frame, or up to 4 reference images, 3 video clips and 3 audio tracks. |
| Character and brand consistency | Kandinsky Lab's report says Pro beats Kandinsky 5.0 Pro and LTX 2.5 in its own human evaluation, with a clear edge in speech quality. Those are the vendor's numbers; with 5-second clips, test whether a character holds across the separate clips a longer scene needs. |
| API, batch, and team | Kandinsky Lab ships code and weights rather than a hosted API, so batch, queueing and access control are yours to build. Imgo exposes batch, team and API surfaces around the models it hosts. |
| Price and free use | The weights are free; the cost is GPU time. The repository's benchmarks put one 5-second clip at about 31 seconds on an H100 with Lite Distill at standard resolution, and about 400 seconds with Pro at Full HD; consumer cards such as an RTX 4090 need offloading and run far slower. Kandinsky 6.0 Video is not priced on Imgo, where every run quotes its credits live before you submit. |
Where Kandinsky 6.0 Video is strong
Kandinsky 6.0 Video comes in two sizes, Lite with 3B parameters and Pro with 29B, plus distilled versions of each that sample in 10 steps instead of 50. Both generate 5-second clips with synchronized 44 kHz audio, including lip-synced speech, from text or from an image plus text. A separate super-resolution model upscales the output to 1920×1080. Code and weights are published under the MIT license, so commercial use and fine-tuning are open.
A reproducible test for your team
Suggested test conditions; no independent benchmark result is claimed.
Prompt
A 5-second 16:9 shot: a street musician in a red scarf plays three notes on a trumpet under a bridge at dusk, lowers it and says one short line to the camera; traffic hums in the background.
Input and controls
Use the same prompt and 16:9 framing on both paths and generate three takes each, at 5 seconds: Kandinsky 6.0 Video Pro with super-resolution on your own GPU, and Seedance 2.0 on Imgo. Keep every take.
Record these outcomes
- The spoken line in sync with the lips
- The three trumpet notes landing on the fingering
- Hands and trumpet intact through the move
- Cost per accepted take, GPU time included
Check live pricing before you decide
Not priced on Imgo
Plans, free usage, API access, and model limits change. Compare the current pages and run the same test batch on both services.
Frequently asked questions
- Can I use Kandinsky 6.0 Video on Imgo?
- No. Kandinsky 6.0 Video is not integrated into Imgo and has no Imgo credit price; this page is an external reference, last checked on 2026-10-23. Seedance 2.0 is the video model to run on Imgo meanwhile.
- Is Kandinsky 6.0 Video open source?
- Yes. Kandinsky Lab published the code and the weights of Lite (3B) and Pro (29B), including distilled versions and the super-resolution model, under the MIT license.
- How long are Kandinsky 6.0 Video clips?
- 5 seconds per generation, with 44 kHz audio generated in the same pass. A separate super-resolution model brings the output to 1920×1080.
- What hardware does Kandinsky 6.0 Video need?
- An NVIDIA GPU. The repository benchmarks cards from a 16 GB RTX 5060 Ti to an 80 GB H100; smaller cards rely on offloading and quantization, and Pro at Full HD took about an hour per clip on the RTX 5060 Ti.