World Labs, under the leadership of Li Feifei, has proudly introduced Atlas—the world's inaugural multimodal world model. This groundbreaking innovation is a multimodal autoregressive diffusion Transformer, boasting a suite of sophisticated capabilities including camera-controlled generation, spatial reconstruction, spatiotemporal simulation, and high-fidelity image generation. Remarkably, Atlas can reconstruct a comprehensive 3D world from a mere single image and craft immersive training environments for robotic systems.

In performance benchmarks, Atlas has demonstrated superior prowess over existing state-of-the-art models in tasks involving camera-controlled generation and intricate 3D reconstruction. Its versatility paves the way for a multitude of applications, spanning from cutting-edge visual effects to advanced robotics, heralding a significant leap forward in the Real-to-Sim domain—a pivotal area for robotics development.

Currently, Atlas is accessible for early adoption by select partners, with opportunities to request usage permissions available through the official website. Looking ahead, Atlas is poised to become the foundational model underpinning World Labs' suite of related products, driving innovation and setting new standards in the field.