Google DeepMind links Street View to Genie for interactive world simulations

Google DeepMind is pushing its world models from invented terrain into mapped reality. By linking Street View with Project Genie, the company is moving beyond fantasy landscapes and synthetic demos toward interactive simulations anchored in real places, a shift that could matter for robotics, autonomy, navigation and other systems that depend on understanding how the world actually behaves.

The effort builds on a rapid run of model releases. DeepMind introduced Genie in February 2024 as an 11-billion-parameter foundation world model trained in an unsupervised way on unlabeled internet videos. The company later described Genie 2 as a large-scale foundation world model that can simulate virtual worlds and the consequences of actions. On August 5, 2025, DeepMind announced Genie 3, a general-purpose model that can generate dynamic worlds navigable in real time at 24 frames per second, keep them consistent for a few minutes and render them at 720p. That progression shows a clear ambition: not just generating scenes, but simulating cause and effect.

Project Genie brought those ideas into a public prototype in March 2026, when Google made it available to Google AI Ultra users in the United States over 18. Google said users can create, explore and remix interactive worlds, with Genie 3 powering the prototype. The company also said the system still has limits, especially around world realism and character control, which underscores how early this remains even as it moves into wider testing.

The Street View connection gives the project a new strategic edge. Street View is not a fictional environment; it is a vast visual record of roads, neighborhoods and public spaces. Folding that into Genie suggests simulations that are grounded in actual geography rather than only imagined settings. For embodied AI, that could mean better training on turns, intersections, weather changes and other conditions that matter to machines moving through the physical world. It also raises the stakes for sectors that prize accurate modeling of terrain and infrastructure, where better world simulation can translate into better decision-making.

DeepMind has said its work on world models rests on more than a decade of research in simulated environments for games and robotics, and it has framed Genie 3 as part of that broader push. The technology now points toward a harder question for regulators and the public: what oversight should apply when a private company can model public space at this level of fidelity, and turn that model into a training ground for machines that may eventually act in the real world.