World Labs Unveils Atlas: A Native Multimodal World Model

mark_k · x · 2026-09-02

World Labs has introduced Atlas, a next-generation world model designed for spatial intelligence.

Key Features

Native Multimodality: Trained from scratch to natively operate on text, images, video, and 3D.
Architecture: A multimodal autoregressive diffusion transformer that combines all inputs into a shared spatial context.
Scalability: Performance improves with increased training compute, a trend expected to continue as the model scales.

Core Capabilities

Camera-Controlled Generation: Generates images and videos with pixel-perfect camera control from one or more input images, outputting up to 1 minute of 1440p video.
Spatial Reconstruction: Reconstructs real-world scenes from one to dozens of input images, generating novel view frames and explicit 3D outputs, outperforming state-of-the-art specialized models.
Space-Time Simulation: Models space and time from input videos, enabling dramatic visual reframing and Real-to-Sim workflows for robotics.
Image Generation: Creates images and 360° panoramas from text, capable of following complex prompts, rendering text, and generating diverse visual styles.

More from Multimodal