Skip to content
Agientry
ZH
← Back to Videos

Lyra 2.0: Explorable Worlds from One Image

Loading video
Loading video

NVIDIA combines camera-controlled video generation with feed-forward 3D reconstruction. Spatial memory retrieves relevant past views, while self-augmented training teaches the model to correct drift, turning one image into persistent, revisitable worlds that can be exported as 3D Gaussians or meshes for robot simulation.

Lyra 2.0NVIDIAGenerative 3DWorld ModelsVideo Generation3D reconstructionRobot simulationEmbodied AI
Category: world-model
Author: @NVIDIAAI
Date: 2026-11-06T00:00:00
Duration: 35.478s
Source: https://x.com/NVIDIAAI/status/2118800021013643269
Reference: https://nv-tlabs.github.io/Project-Lyra/

Related Videos

MUNDO: A Few Photos Become an Interactive, Simulatable World
2:22

MUNDO: A Few Photos Become an Interactive, Simulatable World

@zunnanxu · 162 views · 2026-10-23
MUNDOWorld Models
Programmable World Model: Executable State Separated From Rendering
1:17

Programmable World Model: Executable State Separated From Rendering

@alayastd · 870 views · 2026-09-09
World ModelsVideo Generation
UniAction Unifies Four Action Types in World Models
0:15

UniAction Unifies Four Action Types in World Models

@zizhang_li · 179 views · 2026-10-26
World ModelsUnified Action Conditioning
Juno: One Action-Conditioned JEPA Across the VLA Lifecycle
0:52

Juno: One Action-Conditioned JEPA Across the VLA Lifecycle

@walterzhu8 · 191 views · 2026-10-19
World ModelsJEPA
Multics: Multi-Player World Model Scaling to 10 Players
0:30

Multics: Multi-Player World Model Scaling to 10 Players

@XihuiLiu · 288 views · 2026-10-14
World ModelsVideo Generation
Odyssey 3: World Models Let Robots Rehearse Physics
0:29

Odyssey 3: World Models Let Robots Rehearse Physics

@joaquintenest · 360 views · 2026-10-09
World ModelsSim-to-Real

As an Amazon Associate, we earn from qualifying purchases.