Google DeepMind's Project Genie Turns 280B Street View Images Into Walkable AI Worlds
Google DeepMind connects Project Genie's world model to Street View's 280 billion images, letting anyone simulate and explore real U.S. places as interactive, stylized worlds — now rolling out to AI Ultra subscribers globally.
Takeaways
- Google DeepMind has connected Project Genie to Google Street View, letting users simulate real U.S. locations as interactive, stylized worlds.
- The integration draws on over 280 billion Street View images collected across 110 countries over nearly 20 years.
- Users pick a real U.S. place, choose a visual style (e.g. "Ocean World," "Stone Age"), describe a character, and explore a generated environment in real time.
- Rolling out now to Google AI Ultra subscribers globally ($200/month), ages 18+; Street View grounding currently limited to U.S. locations.
- The primary research use case is robotics and agent training — Waymo already uses Genie to simulate rare driving scenarios like tornadoes and animal crossings.
- Current limits: no physics simulation, video-game-level fidelity, and sessions capped at 60 seconds at 720p/24fps.
Project Genie, Google DeepMind's experimental world-model prototype that lets you build and walk through AI-generated environments, just got a significant upgrade: it can now pull real-world imagery directly from Google Street View. Google DeepMind is connecting Street View to Project Genie, the company's general-purpose world model that can generate diverse, interactive environments. The result is something genuinely new , not a game, not a map, but a navigable simulation of a real place that you can reimagine from scratch.
The integration, announced at the Google I/O developer conference, marks one of the most tangible demonstrations yet of what generative world models can do when paired with a colossal real-world dataset. And the dataset in question is enormous: Project Genie can now draw on more than 280 billion images captured across 110 countries and all seven continents.
What exactly is Project Genie?
A quick primer: a world model is an AI system that doesn't just generate a static image or video clip , it builds a persistent, navigable environment that responds to your movement in real time. Think of it as the model holding a mental map of a space and rendering it as you explore, rather than pre-computing a fixed video. Project Genie is Google's AI-powered system for creating explorable snow-globe environments from written prompts, with creations lasting 60 seconds at 720p and 24 fps.
Users are able to create contained worlds in whatever style they'd like, complete with a character of their own description, and then move a camera around that space. Before this update, those worlds were entirely fictional. Now they can be anchored to a real address.
How the Street View grounding works
The feature is powered by Maps Imagery Grounding, which Google says developers use to create AI visuals with Street View. In practice, the workflow is straightforward:
- Open Project Genie in Google Labs and tap the Maps pin.
- Choose a real place in the U.S. from Google Maps.
- Pick a visual style , options include "Desert Sands," "Stone Age," "Ocean World," and "B&W Film," among others.
- Describe your character (an animal, a comic book hero, a claymation monster).
- Genie generates an interactive world with its starting location tied to Street View imagery.
The official blog post gives a vivid example: select "Ocean World" to scuba dive with fish around the Golden Gate Bridge, or pick "B&W Film" to see the Fort Worth Stockyards as they might have looked in the 1920s, complete with saloons and vintage cars.
What makes this technically interesting is the spatial continuity. Jonathan Herbert, principal product manager for Maps, pointed out that the real breakthrough isn't photorealism , it's that if you turn 360 degrees, the model correctly remembers and simulates the environment behind you. From that anchor point, it builds a new world on top of real geography. That's a harder problem than it sounds for a generative model.
The research angle is the bigger story
The consumer demo is fun, but the deeper motivation is robotics and agent training. "It's really powerful for both the agent [and robotics] use case and for humans to play with, and that's always been the thesis of Genie," Jack Parker-Holder, a research scientist on DeepMind's open-endedness team, told TechCrunch.
The concrete example Parker-Holder gave: a robot deployed in London rarely encounters direct sunlight. Genie can simulate those rare moments , sunlight glinting off Victorian housing , so the robot isn't caught off guard when it happens in the real world. Waymo already uses Genie to train robotaxis on rare scenarios such as tornadoes and animal crossings. Adding Street View data to that pipeline could help Waymo prepare to launch in cities where it has no prior sensor data.
There's also a key difference from Waymo's existing simulators: those are built entirely from the car's point of view. Street View allows Genie to shift perspective to other agents , a pedestrian, a delivery robot, a humanoid , navigating the same physical space.
Where it falls short (for now)
Google is being transparent about the limitations, and they're real:
- No physics. The models are not yet physics-aware. Jack Parker-Holder of DeepMind estimates a six to twelve month gap compared to pure video models. In demo footage, a character running through a snowy Joshua Tree simply passes through cacti and bushes.
- Not photorealistic. Results are described as impressive and recognizable, but video-game quality rather than faithful reconstructions of real streets.
- U.S. only for now. Street View imagery in Project Genie is available now for places in the U.S., with plans to expand over time.
- Still experimental. Project Genie is still an experimental research prototype in Google Labs, so Google is working behind the scenes to make the details even sharper and more accurate.
Who can use it and what it costs
Project Genie, including the Street View capability, is gradually rolling out to all eligible Google AI Ultra $200 subscribers globally, age 18 and older. That's the same $200/month tier that unlocks Google's other frontier AI features. The Street View grounding is included , no additional cost on top of the subscription.
The broader significance here is what this integration signals about the direction of world models. For years, the argument for world models in AI research has been theoretical: simulate enough of reality and you can train agents cheaply, at scale, without physical hardware. Google DeepMind has connected its Project Genie world model to 20 years of Street View imagery, allowing users to wander through AI-generated simulations of real places. That's the theory becoming practice , and the 280 billion image corpus is a moat that's very hard to replicate.