Kandinsky LabKandinsky VSR
Kandinsky 6.0 VSR
This video super-resolution release works on existing frames. Its package includes the trained upscaling transformer, a video autoencoder and learned latent upscalers.
Upscale frames in overlapping tiles
The pipeline supports 2×, 2.25× and 4× scaling. It blends overlapping regions after decoding. Its video-only output needs the original audio attached again when sound must be retained.kandinskylab/Kandinsky-6.0-VSR-5s-Diffusers — model card and files
Separate the transformer size from the package
The bundle also lists a 1.74B video VAE and a 3.65B bank of 2× and 4× latent upscalers. The transformer count is not the total pipeline size.kandinskylab/Kandinsky-6.0-VSR-5s-Diffusers — model card and files
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | kandinskylab/Kandinsky-6.0-VSR-5s-Diffusers |
| SR transformer | 1.41B parameterskandinskylab/Kandinsky-6.0-VSR-5s-Diffusers — model card and files |
| Sampling | Four Euler steps per tilekandinskylab/Kandinsky-6.0-VSR-5s-Diffusers — model card and files |
| Weight license | MITkandinskylab/Kandinsky-6.0-VSR-5s-Diffusers — model card and files |