World Labs Unveils Atlas, an Omni World Model That Reshoots Videos From Camera Angles That Never Existed

FinanceYF5 · x · 2026-09-02

World Labs has introduced Atlas, its next-generation omni world model for spatial intelligence. Pretrained from scratch, it natively operates on text, images, video, and 3D, using a multimodal autoregressive diffusion transformer that combines all inputs into a shared spatial context while staying consistent in 3D.

Key capabilities:

Atlas is built to scale, with performance improving as training compute grows, and will power future versions of Marble. In the showcased demo, footage from just three tripod-mounted phones is enough for Atlas to reconstruct the full scene and reframe it from camera positions that never existed — video generation is moving from editing pixels to genuinely understanding 3D space.

Related event: World Labs Unveils Atlas, an Omni World Model for Spatial Intelligence(53 posts)→

Original post →

More from Multimodal

Multimodal channel →