Kandinsky LabKVAE
KVAE Image 1.0
This image autoencoder converts RGB pictures into continuous latents and reconstructs them. It is the original image tokenizer in the KVAE family.
Research release
Use the latent representation as a model component
The image card reports reconstruction on ImageNet-256 and DIV2K. Its 16-channel latent representation is the interface a downstream generator must use.kandinskylab/KVAE-2D-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-2D-1.0 |
| Compression | 8 × 8 spatialkandinskylab/KVAE-2D-1.0 — model card and files |
| Latent channels | 16kandinskylab/KVAE-2D-1.0 — model card and files |
| Weight license | MITkandinskylab/KVAE-2D-1.0 — model card and files |