World Labs' Atlas turns 3 photos into a full 3D world, 50-100x cheaper to capture
drfeifei · x · 2026-09-05
In an a16z conversation, World Labs founders Fei-Fei Li, Justin Johnson, and Ben Mildenhall join Martin Casado to discuss Atlas, a world model for spatial intelligence.
- New paradigm: LLMs use next token prediction, video models use next frame prediction — Atlas is built on next view prediction, and is the first model to unify pixel generation and pixel reconstruction, two tracks computer vision kept separate for half a century.
- Practical result: a 50-100x reduction in the cost of digitally capturing 3D representations — previously 100-300 photos per room, now just 3.
- The team frames the journey from a 2012 ray tracing course blog to co-founding World Labs in 2024 as "curiosity compounds."
More from Multimodal
- Kling joins WPP and Adobe at Advertising Week NY on AI video's first real production beachhead — lmoroney · 2026-10-18
- Midjourney style code --sref 3689510055 produces moody dark-green cinematic landscapes — sergeantsref · 2026-10-18
- AI-generated Instagram influencer hits 57K followers as Redditors hunt for its video workflow — Great_Scale_2925 · 2026-10-18
- Apple's immersive horror short Remnants wows Vision Pro users with 3D presence — Scobleizer · 2026-10-18
- H3 360° Orbit Camera LoRA for MiniMax Open-Sourced, Generates Smooth 5.2s Shots — merica420_69 · 2026-10-18
- 'Tesla in Pakistan' AI video made with Seedance 2.5 goes viral — SimplyAnnisa · 2026-10-18