Google DeepMind · 2026-05-19 · major
Project Genie Plugs Into Street View — Google's World Model Generates Interactive Stylised Scenes Anchored to Real US Locations, Rolling Out to AI Ultra Subscribers Globally
Project Genie can now ingest Google Maps Street View imagery as a scene seed: pick a US address, apply a style (Ocean World, B&W Film), and walk around a generative version of that exact street. Available today to $200 AI Ultra subscribers.
Google's Genie world model now takes a real Street View address as its scene seed — pick a styled lens, walk around a generated version of that exact block.
Key specs
| Subscription tier | Google AI Ultra ($200/mo) |
|---|---|
| Initial coverage | US Street View locations |
| Age gate | 18+ |
| Prior use | Agent training, Waymo autonomous-vehicle simulation |
What is it?
Project Genie is Google DeepMind's general-purpose world model: a system that takes a prompt or seed and generates an interactive 3D environment a user (or an AI agent) can walk through and manipulate. The May 19, 2026 update connects Genie to Google Maps Street View. You pick a US address inside the Genie experience, choose a creative style — examples in Google's post include 'Ocean World' (the location reimagined underwater) and 'B&W Film' — and Genie generates an interactive scene anchored to the real-world geometry of that block. Until now Genie's main consumers were Waymo's autonomous-driving simulator and DeepMind's agent-training pipeline.
How does it work?
Genie pulls Street View panoramas, lifts them into a navigable 3D scene, and rewrites the appearance via a style prompt while preserving the geometry of the real location. Users move through the generated world rather than just viewing a single frame, so it functions as a playable scene rather than a 2D image. Style transfer and content generation happen entirely server-side; Google is gating it behind the AI Ultra tier to manage compute. The blog post explicitly calls out that Genie is still a research model and access is being rolled out gradually to all eligible Ultra subscribers globally as part of the I/O 2026 wave.
Why does it matter?
World models that can ingest a real location are the missing piece between generative video (Veo, Gemini Omni) and embodied AI training (Waymo, robotics). Anchoring to Street View makes it possible to generate consistent training data for places that actually exist, which is a more useful substrate for agent training, navigation research and real-world simulation than free-form text-to-3D. For consumers, it's also Google's most concrete answer to 'what does a world model do for me' — pick your block, see it underwater.
Who is it for?
AI Ultra subscribers curious about generative worlds, robotics/agent researchers needing geographically grounded scenes, Waymo-style simulation teams.
Try it
Open the Project Genie experience in your Google AI Ultra account, search for a US Street View location, and pick a style — US locations only at launch.