KyutaiCADRE
CADRE 4x16x16-multires
This CADRE autoencoder compresses video into a representation for interactive world models. One latent position covers 4 frames and a 16-by-16-pixel region, with 64 channels.
Research release
The codec reconstructs video from compressed features
The final training phase uses five image sizes from 256 by 448 to 384 by 640. Preprocessing keeps a clip’s size and pads to the compression grid. This differs from the original fixed-resolution checkpoint.cadre-4x16x16-multires: Kyutai model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | kyutai/cadre-4x16x16-multires |
| Terms | CC BY-NC 4.0 weights, with DINOv3 terms for the frozen encoder.cadre-4x16x16-multires: Kyutai model card |
| Training stage | 400,000 steps, including a final 40,000-step multi-resolution phase.cadre-4x16x16-multires: Kyutai model card |
Related models
- CADRE 2x16x16 world models
Kyutai · CADRE
Four separately trained world models that predict video from context frames and actions in distinct environments.
Launch date not confirmedResearch releaseDownloadable