MongoDB - Build AI That Scales
Mini toolsLyra 2.0: Explorable Generative 3D Worlds
By NVIDIA
Lyra 2.0 is NVIDIA’s framework for building explorable generative 3D worlds at scale. It starts from an input image, follows a user-defined camera trajectory with an optional text prompt, iteratively generates video segments, and reconstructs them into 3D point clouds that feed back into continued navigation. The system is designed to solve two major long-horizon failure modes, spatial forgetting and temporal drifting, using per-frame 3D geometry for information routing and self-augmented training to teach drift correction. NVIDIA also shows export to 3D Gaussian splats, meshes, and Isaac Sim for physically grounded robot navigation and interaction.
New
Multimodal
Gen 3
Overview
Lyra 2.0 is NVIDIA’s generative 3D world model for creating persistent, explorable virtual environments from an input image. It generates camera-controlled walkthrough video and lifts it into 3D, enabling long-horizon, 3D-consistent world generation that can be exported as Gaussian splats or meshes for rendering, simulation, and embodied AI use.
About NVIDIA
Industry:
Computer Hardware Manufacturing
Company Size:
42000
Location:
Santa Clara, California, US
Website:
nvidia.com
Tools using Lyra 2.0: Explorable Generative 3D Worlds
No tools found for this model yet.
Last updated: April 15, 2026
