World Labs unveils Atlas, an omni world model for spatial intelligence
pstAsiatech · x · 2026-09-06
World Labs (Fei-Fei Li's startup) has introduced Atlas, a next-generation world model pretrained from scratch to natively operate on text, images, video, and 3D. It's a multimodal autoregressive diffusion transformer that keeps a shared spatial context, staying 3D-consistent and imagining beyond what it has seen.
Key capabilities:
- Camera-controlled generation: pixel-perfect camera control from one or more images, up to 1 minute of 1440p video.
- Spatial reconstruction: reconstructs real scenes from 1–dozens of images, beating specialized SOTA 3D reconstruction models.
- Space-time simulation: reframes video and enables Real-to-Sim workflows for robotics.
- Image generation: text-to-image and 360 panoramas with complex prompts and text rendering.
Performance scales with training compute; Atlas will power future versions of Marble.
More from Embodied
- Asimov humanoid robot Day 345: testing full-body teleoperation — freelerobot · 2026-09-06
- First-time Tesla Cybercab robotaxi rider loves the experience, says whurley — whurley · 2026-09-06
- Polymarket Puts Only 22% Odds on Tesla Launching California Robotaxis by Year-End — Polymarket · 2026-09-06
- Paraplegic user rates Tesla CyberCab 10/10 in viral accessibility review — elonmusk · 2026-09-06
- HKUST(GZ) lab lands 3 CoRL 2026 papers, unveils terrain-adaptive robot motion model — Scobleizer · 2026-09-06
- Musk signals Tesla could turn Cybercab tech into a consumer AI-driven car lineup — elonmusk · 2026-09-06