TLDRocket
Sign in

Simulate real-world places with Project Genie and Street View

Google DeepMind

Google's Project Genie can now build interactive virtual worlds anchored to real Street View locations. Ultra subscribers can scuba dive under the Golden Gate Bridge or see 1920s Fort Worth, all AI-generated from real map data.

Based on reporting by Google DeepMind — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

Google DeepMind's Genie has spent the past year or so quietly becoming infrastructure — the kind of thing researchers use to train agents in simulated worlds, and Waymo leans on to model realistic roads. Now Genie is getting a dose of geography. Google is wiring its world-generation model into Street View imagery, so instead of dreaming up entirely fictional landscapes, Genie can start from a real place and warp it into something else.

The mechanics are simple enough. Inside Project Genie, you tap a Maps pin to pick a U.S. location, choose a visual style — Desert Sands, Stone Age, B&W Film, that sort of thing — and describe a character to drop into the scene. Genie then builds an explorable, interactive world rooted in that spot's actual Street View data. Want the Golden Gate Bridge submerged, with fish schooling around its towers? There's an Ocean World preset for that. Curious what the Fort Worth Stockyards looked like a century ago, saloons and all? Switch to black-and-white film and it reimagines the same physical location in period detail.

Under the hood this runs on Maps Imagery Grounding, the same tech stack developers already use to generate polished AI visuals from Street View photos. Diego Rivas, Jonathan Herbert, and Nicole Segaran — the trio leading this across DeepMind, Maps, and Labs — frame it as Genie 'anchoring itself in reality,' which is a fair way to describe the shift from pure imagination engine to something closer to a real-world simulation sandbox. That distinction matters beyond novelty apps: a model that can generate plausible, interactive environments grounded in actual streets is exactly the kind of training ground robotics and autonomous-vehicle teams want for agents that eventually have to operate outdoors.

Coverage is U.S.-only for now, with Google promising expansion later, and the tool remains explicitly experimental — DeepMind is upfront that accuracy and detail still need work. Access is also gated behind Google AI Ultra, the $200-a-month tier, which is rolling out globally to subscribers 18 and older starting now. So this isn't a mass-market toy yet. It's a preview of where Genie is headed, wrapped in a feature built for people already paying Google's steepest AI subscription price.

My take — AI-written commentary, not fact-checked reporting

Grounding a world model in real Street View data is the more interesting story here than the scuba-diving Golden Gate Bridge demo — this is Google quietly building a simulation layer for robots and driving systems using imagery it already owns outright, and nobody else has. That's the actual moat, not the $200 subscription gimmick sitting on top of it.

Read more about this at: Google DeepMind

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.