GAIA-2

Overview

GAIA-2 is a cutting-edge video-generative world model for safer assisted and automated driving. It generates diverse, controllable, and safety-critical driving scenarios at scale, covering multiple geographies (UK, US, Germany) and conditions. It supports multi-camera viewpoints, fine-grained control over ego-action, environment, and road configuration, and can synthesize rare edge cases. GAIA-2 augments real-world data and enables action-conditioned generation, accelerating validation of autonomous systems.

Key features

• ▸Video-generative world model for autonomous driving
• ▸Fine-grained control over ego-action, environment, and road configuration
• ▸Multi-camera viewpoint generation with spatiotemporal consistency
• ▸Geographic coverage: UK, US, Germany
• ▸Supports long-horizon stable video synthesis
• ▸Generates safety-critical scenarios (e.g., cut-ins, emergency braking)
• ▸Out-of-distribution generalization capability
• ▸Action-conditioned generation for systematic validation