Spatial Intelligence: Atlas, World Model from World Labs

World Labs has introduced Atlas, an omni world model designed for spatial intelligence that natively processes text, images, video, and 3D data. Built as a multimodal autoregressive diffusion transformer, Atlas grounds inputs in a shared 3D spatial context to generate, reconstruct, and simulate environments.

Core Capabilities

• Camera-Controlled Generation: Generates up to 1-minute videos at 1440p resolution from reference images with precise camera movement and 3D consistency.
  
  
• Spatial Reconstruction: Reconstructs scenes from sparse input images (from 1 to dozens), outputting novel views, point clouds, and 3D Gaussian splats while outperforming specialized 3D reconstruction models.
  
  
• Space-Time Simulation: Enables video reframing ("bullet time" effects) from simple camera setups and supports Real-to-Sim workflows for robotics training and evaluation.
  
  
• Image Generation: Produces high-fidelity images and 360-degree panoramas from text prompts with detailed style and text rendering capabilities.
  
  

Technical Architecture & Performance

• Model Type: Combines autoregressive sequence modeling with rectified flow diffusion, leveraging transformer-based scaling for continuous compute improvements.
  
  
• Performance: Outperforms specialized baselines in human evaluation for camera control and achieves lower relative error across standard 3D reconstruction benchmarks (e.g., DTU, ETH3D, ScanNet).
  
  

Atlas is not open source. It is a proprietary, cloud-accessible model being integrated into World Labs' platform products like Marble.

Currently, it functions as a commercial preview/announcement:

• Access: It is not open to the general public yet. World Labs is accepting requests for early access for select partners.

• Deployment: Once rolled out, it will power World Labs' cloud-based creation platform (Marble) and be offered via enterprise cloud APIs rather than downloadable local model weights.

Atlas: A World Model for Spatial Intelligence | World Labs

similar: 

Physical AI with World Foundation Models | NVIDIA Cosmos

New iOS26 Spatial Scenes - Turn Any Photo into a 3D Time Portal! - YouTube

Capture, view, and share spatial photos and videos on Apple Vision Pro - Apple Support

Amazon.com: OPIC Smartphone 3D Stereoscopic Lens 3D Camera Stereo Photos Device for Phone, Tablet, Laptop – OPIC Spatial Clips on to Turn Your Device Into a 3D Camera for Photos, Videos and VR-Ready Memories