World Labs unveils Atlas, an omni world model pretrained from scratch for spatial intelligence
drfeifei · x · 2026-09-02
World Labs introduces Atlas, its next-generation omni world model for spatial intelligence:
- Architecture: a multimodal autoregressive diffusion transformer pretrained from scratch to natively operate on text, images, video, and 3D. All inputs combine into a shared spatial context, keeping generation 3D-consistent; performance scales with training compute.
- Four capability areas: camera-controlled generation (up to 1-minute 1440p video with pixel-perfect control); spatial reconstruction (1 to dozens of images, outputs novel views and explicit 3D, beating specialized 3D reconstruction SOTA); space-time simulation (video reframing, Real-to-Sim robotics workflows); and image generation (text-to-image and 360 panoramas with text rendering).
Atlas will power future versions of Marble and other products.
Related event: World Labs Unveils Atlas, a Pixel-Perfect Multimodal World Model(22 posts)→
More from Multimodal
- Meta Avatars 2.0: Stylized FACS Implementation Details — SergiCaballer · 2026-09-02
- Book 'KI-KUNST' explores the creativity and controversy of AI art — Merzmensch · 2026-09-02
- Interactive brain model: AI traces anatomy when you hear 'pass the salt' — mikeyk · 2026-09-02
- AI-Generated Drink Ad Features Eye Reflections and Splashes — anthara_ai · 2026-09-02
- Prompt for Fabric and Light Title Sequence via MiniMax H3 — umesh_ai · 2026-09-02
- Interest shifts to Meta's real-time voice transcription model — IndraVahan · 2026-09-02