Google DeepMind has added a sharper real-world framing to Genie 3, its interactive world model: the company now says Genie is grounded in Google Maps Street View data, allowing users to create unexpected worlds anchored in reality.

The claim matters because Genie 3 is not only a video generator. DeepMind describes it as a general-purpose world model that generates photorealistic environments from text and lets a user or agent move through them in real time. The model page lists 20 to 24 frames per second interaction, 720p output, world consistency when revisiting areas, and Street View grounding as core capabilities.

DeepMind's earlier technical announcement positioned Genie 3 as a research step toward richer simulation environments for AI agents. It said the model can produce dynamic worlds that remain consistent for a few minutes, giving agents a place to test goals and actions without a hand-built game engine or simulator.

The Street View note does not mean Genie can recreate a specific address accurately. DeepMind's own limitations section says exact representation of real-world locations is still an open problem, and that the model supports only a few minutes of continuous interaction rather than long sessions.

That conservative reading is the important one: Genie 3 is moving world models closer to real-world simulation, but today's version is better understood as a controllable, reality-informed generator than a faithful digital twin.