Genie 3

Genie 3 is a world model from Google DeepMind, announced in August 2025, that generates interactive 3D environments from a text prompt and renders them in real time at 720p and 24 frames per second, holding a scene consistent for a few minutes. This foundation model represents a significant evolution in AI-driven world generation, enabling users to interact with dynamically created environments through learned action spaces. Genie 3 demonstrates sophisticated understanding of physics, object relationships, and environmental dynamics while generating diverse interactive scenarios from minimal input data. The model excels at creating game-like environments, simulations, and virtual worlds that respond realistically to user inputs.

Built on transformer architecture with enhanced spatiotemporal reasoning capabilities, Genie 3 serves as a powerful tool for AI research, game development, and interactive content creation, supporting promptable world events that let a user change the weather or introduce objects mid-session, and DeepMind has used it to test its SIMA agent, which pursued goals inside the generated worlds.

Related services: Agentic AI consulting.

← Back to the AI Glossary