World Labs introduced Atlas as the first multimodal world model that generates image and video frames with pixel-perfect camera control and reconstructs them in 3D. Pretrained from scratch as a multimodal autoregressive diffusion transformer, it turns a few photos into explorable spaces, 1440p camera-controlled video, and photorealistic RGB-D for robot simulation. Early access opens in the coming weeks.

Key Takeaways

  • βœ“One architecture covers generation, explicit 3D reconstruction, and space-time simulation
  • βœ“A handful of ordinary cameras can become a bullet-time studio or a robot RGB-D simulator
  • βœ“Atlas beats top open-source reconstruction models; early access is opening soon
ADSponsored