Kandinsky LabKVAE
KVAE Video 2.0 t4s16
This causal video autoencoder converts frame sequences into continuous latents. The 2.0 release supports long sequences through temporal caching.
Research release
Use the latent representation as a model component
The card evaluates reconstruction at 720p. Keep each video stream on its own model instance because internal temporal caches cannot be interleaved across independent streams.kandinskylab/KVAE-3D-2.0-t4s16 — 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-2.0-t4s16 |
| Compression | 4 × 16 × 16 temporal/spatialkandinskylab/KVAE-3D-2.0-t4s16 — model card and files |
| Latent channels | 64kandinskylab/KVAE-3D-2.0-t4s16 — model card and files |
| Weight license | Apache 2.0kandinskylab/KVAE-3D-2.0-t4s16 — model card and files |