Skip to content
Agientry
← Back to Videos

CADRE: Real-Time Video Autoencoders for World Models

Loading video
Loading video

Kyutai releases CADRE, an open causal diffusion video autoencoder that lets world models respond to actions at 20 fps with under 230 ms lag on a single GPU.

WorldModelVideoAutoencoderDiffusion modelsRealTimeRobotManipulationKyutaiWorld Models
Category: research
Author: @kyutai_labs
Date: 2026-10-22T00:00:00
Duration: 9.6s
Source: https://x.com/kyutai_labs/status/2113262102924476502
Reference: https://arxiv.org/abs/2610.30129

Related Videos

WorldKV: retrieval and compression for world-model memory
0:14

WorldKV: retrieval and compression for world-model memory

@jungyi3578 · 319 views · 2026-05-23
WorldModelKVCache
LongCat-Video: Meituan open-sources a 13.6B video model
0:30

LongCat-Video: Meituan open-sources a 13.6B video model

@denziideng · 360 views · 2026-10-06
VideoGenerationText-to-Video
RoFacto: Action-Conditioned Diffusion World Model
0:48

RoFacto: Action-Conditioned Diffusion World Model

@byungjun__kim · 421 views · 2026-07-27
RoFactoWorld Models
RoboJEPA: The First Scaling Law for Multi-Embodiment World Models
1:06

RoboJEPA: The First Scaling Law for Multi-Embodiment World Models

@artemZholus · 264 views · 2026-10-10
World ModelsJEPA
JEPA-Anything: One World Model Across Six Domains
0:30

JEPA-Anything: One World Model Across Six Domains

@Charles_Y_Wu · 420 views · 2026-10-05
World ModelsEmbodied AI
GAE: A Geometry-Native Latent Space for 3D-Consistent World Generation
0:51

GAE: A Geometry-Native Latent Space for 3D-Consistent World Generation

@HuggingPapers · 777 views · 2026-09-23
World Models3D Generation

As an Amazon Associate, we earn from qualifying purchases.