Kandinsky LabKandinsky 5.0 Video
Kandinsky 5.0 T2V Lite Pretrain 10s
Generate ten-second video from text with this 2B Lite checkpoint. It is the pretraining release for further adaptation.
Research release
Use the pretraining checkpoint for adaptation
The model zoo distinguishes pretraining weights from SFT and distilled weights. The ten-second line uses NABLA sparse attention.kandinskylab/Kandinsky-5.0-T2V-Lite-pretrain-10s — model card and files
Account for the complete video pipeline
The generator is used with Qwen2.5-VL and CLIP text encoders and a HunyuanVideo VAE. Those upstream components are not new Kandinsky foundation models.kandinskylab/Kandinsky-5.0-T2V-Lite-pretrain-10s — model card and files
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | kandinskylab/Kandinsky-5.0-T2V-Lite-pretrain-10skandinskylab/Kandinsky-5.0-T2V-Lite-pretrain-10s-Diffusers |
| Generator size | 2B parameterskandinskylab/Kandinsky-5.0-T2V-Lite-pretrain-10s — model card and files |
| Weight license | MITkandinskylab/Kandinsky-5.0-T2V-Lite-pretrain-10s — model card and files |
| Clip duration | Up to 10 secondskandinskylab/Kandinsky-5.0-T2V-Lite-pretrain-10s — model card and files |