Starting today, Google DeepMind connects Street View data to Project Genie, letting some U.S. Ultra subscribers simulate real places in an interactive way—adjusting conditions like weather—while researchers warn the results are still closer to video games than
The little person icon has long been a fun trick on Google Maps. But today, Google is pushing that idea further—into something closer to stepping onto a street and changing what happens next.
Starting today. Google DeepMind is connecting Street View to Project Genie. the company’s general-purpose world model designed to generate diverse. interactive environments. The rollout began during the Google I/O developer conference. and it begins with some Ultra users in the United States. with access expanding at scale over time.
For Jack Parker-Holder. a research scientist on DeepMind’s open-endedness team. the appeal is simple: Genie is built for both agent and robotics use cases. but also for humans to play with. He described how the model could help when robots face rare conditions—like sunlight in London. where the sun “rarely sees” daylight. Genie. Parker-Holder said. could simulate those scarce moments when light glints off London’s Victorian housing so a robot isn’t startled by sudden changes.
He also framed Genie as a way to “see” conditions before they arrive. “Simultaneously, you might say, ‘I’m going to New York City, but not this time of year,’” he said. “‘It’s going to be snowy. I want to see what that block looks like in the snow.’”
Google’s bet is that Street View supplies the grounding—20 years of imagery gathered through cars with cameras and individuals wearing “tracker backpacks.” The company says it has collected north of 280 billion images across 110 countries and seven continents. building a huge real-world reference library.
“With Street View, we have imagery from a large quantity of the world,” Parker-Holder said. “You can imagine how potentially powerful it is to combine this rich source of real-world information and data with an ability to simulate worlds.”
This isn’t the first time Google has put Genie into the hands of paying users, either. Genie 3 was released for research preview last August. and in January Google opened access to the tool to Google AI Ultra subscribers in the U.S. allowing customers to create interactive game worlds from text prompts or images. The stated goal for Genie includes educational experiences, gaming, and robotics training.
Waymo is already a proving ground. Genie 3 helps power a Waymo simulator used to train self-driving cars on “exceedingly rare events. ” including tornadoes or “casual elephant encounters.” Adding Street View data could help that kind of training as Waymo prepares to launch in more cities around the globe.
Parker-Holder drew a distinction between existing simulations and what Street View could add. He said Waymo’s simulator scenarios are all from the car’s point of view. With Genie plus Street View. the model can simulate a world anchored to a real place and shift the point of view to other kinds of agents—like a human or a robot.
The Human appeal is clear in the demos. In samples shown by Google’s team—including an underwater simulation of a neighborhood the reporter said they used to live in—the results were described as impressive and recognizable. But they still carry the look of video game output rather than photorealism.
There’s also a technical gap. The models are not yet physics-aware, meaning they don’t understand cause and effect in a reliable way. In one simulation—a woman running through a snowy Joshua Tree—she ran through cacti and bushes.
That difference matters because Google isn’t treating Genie’s Street View integration as a finished product. Diego Rivas, a product manager at DeepMind, said the capability is still an experiment and there’s “much to improve upon” in terms of accuracy.
Google’s broader generative toolchain includes other models with different strengths, and the gap was highlighted through comparisons. The article points to Nano Banana. which can generate perfect text in infographics. and Veo. a video generator that can understand paper boats drifting on water currents. smoke dispersing into the air. and fabric draping over forms.
Physics, in these systems, isn’t “hard-coded,” the report says. Instead, they learn it over time through passive observation, the way a living being would. Parker-Holder suggested the timing gap for models like this: “I think for this kind of model. it’s maybe six to 12 months behind video in terms of the accuracy and quality. ” he said. “So I think it’s something we will solve.”.
Not everyone is talking about realism as the main breakthrough. Jonathan Herbert, director of Google Maps and a Street View team veteran who started as an intern 12 years ago, said Genie can’t yet produce a faithful reconstruction of a street. In his view, the bigger leap is spatial continuity.
Herbert explained that if you turn 360 degrees, the AI correctly remembers and simulates what’s behind you. From there, the model can build a new environment on top of that.
“We have long thought about how we can build out the best and richest model of the world on top of Street View data,” Herbert said. “It’s definitely been an idea of ours to use Maps Data in new ways and for new kinds of AI research for a pretty long time.”
Access is rolling out in the United States first for some Ultra users starting today. with global Ultra users set to get it over the next few weeks. Google is effectively asking people to try the experiment—weather. seasons. and agent viewpoints included—while it works on making the simulations more accurate and more faithful to the real world.
Google DeepMind Project Genie Street View Google Maps AI Ultra Waymo simulator robotics training world model generative AI