Kandinsky LabKVAE
KVAE Video 1.0
This causal video autoencoder converts frame sequences into continuous latents. The 1.0 release is the earlier video tokenizer.
Research release
Use the latent representation as a model component
The original reconstruction evaluation uses 540p clips. The card notes limitations at higher resolutions; this is not a claim of a hard 540p input limit.kandinskylab/KVAE-3D-1.0 — model card and files
Keep the encoder and decoder paired
Practical guidance. A compressed representation is useful only with a compatible consumer and decoder. Replacing an autoencoder can change the latent distribution even when tensor dimensions match.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | kandinskylab/KVAE-3D-1.0 |
| Compression | 4 × 8 × 8 temporal/spatialkandinskylab/KVAE-3D-1.0 — model card and files |
| Latent channels | 16kandinskylab/KVAE-3D-1.0 — model card and files |
| Weight license | MITkandinskylab/KVAE-3D-1.0 — model card and files |