Kyutai CADRE: Video Autoencoders That Cut World Model Lag Below 230 ms
MarkTechPost Michal Sutter
Kyutai released CADRE, an open set of causal diffusion video autoencoders for interactive world models designed to reduce action lag. CADRE world models run at 20 fps on 1 H100 and react within 230 ms. Developers can use MIT-licensed code and non-commercial (CC BY-NC 4.0) CADRE weights, which change world-model video decoding trade-offs toward lower latency and improved driving motion scores (e.g., lower driving FVD).
Why it matters
Kyutai released CADRE, a family of 7 open causal diffusion video autoencoders for interactive world models. World models built on CADRE run at 20 fps on 1 H100 and react in under 230 ms. The post Kyutai CADRE: Video Autoencoders That Cut World Model Lag Below 230 ms appeared first on MarkTechPost.
Related stories
Meet Kyutai REKI: The 2.4B VLA Model Bringing Efficient Memory to Robot Manipulation
MarkTechPost · 6 days ago ·
39
State of open video generation models in Diffusers
Hugging Face · 1 year ago ·
48
The Video Production Stack Now Fits on One Desk: LTX-2.5 Launches as NVIDIA-Accelerated Open Weights World Model
MarkTechPost · 2 months ago ·
21