SAN FRANCISCO, Jan 30, 2026, 07:14 PST
- Google started releasing Project Genie, an interactive world-building prototype, to AI Ultra subscribers across the United States.
- DeepMind’s Genie 3 world model, paired with Nano Banana Pro and Gemini, powers the tool to convert text or images into interactive scenes.
- Google limits sessions to 60 seconds while it experiments with controls, realism, and content safeguards.
On Thursday, Google started rolling out Project Genie, a Google Labs prototype available to U.S. Google AI Ultra subscribers. This new tool allows users 18 and older to generate and explore interactive worlds from text prompts or images, the company confirmed. 1
This launch is significant because Google is moving “world models” from research labs into a paid, consumer-facing experiment. The company is counting on user feedback to refine a technology it believes has applications far beyond gaming.
World models are AI systems designed to predict how an environment evolves as you navigate it—essentially a step toward agents capable of planning actions within a changing scene. DeepMind has made the case that this ability is crucial for AGI, shorthand for a system built to tackle a wide range of tasks instead of just one specific job.
Google says Project Genie runs on DeepMind’s Genie 3 and links up with its Nano Banana Pro image generator and Gemini. It kicks off with “world sketching,” where users either describe a scene and main character or upload an image to get started.
Users choose their mode of movement—walking, riding, flying, or driving—and select either a first- or third-person perspective before the world generates. Google also offers the option to remix existing creations and download videos of these journeys.
Google set a 60-second limit on world generation and navigation. A spokesperson explained the company discovered that this timeframe helps deliver “high quality and consistent” results while allowing more users access to the tool. 2
In a hands-on, The Verge reported the generated worlds ran at around 720p and hovered near 24 frames per second, with movement handled by standard keyboard keys. They also noted input lag that gave some sessions a cloud gaming vibe rather than the feel of a local game.
Diego Rivas, product manager at Google DeepMind, explained the release aims “to learn about new use cases” for the tech, such as visualizing scenes for filmmaking and interactive educational media. He also mentioned more basic tests, like starting a world from a photo of a child’s toy.
Shlomi Fruchter, a research director at DeepMind, warned that Project Genie isn’t an end-to-end product meant for daily use. The company calls it a research prototype and noted that some features mentioned during last year’s Genie 3 discussions haven’t been added yet.
The rollout arrives amid a surge of companies developing their own world models. Startups such as Fei-Fei Li’s World Labs and AI video specialist Runway are also active in this space, TechCrunch reported. 3
But even in its early release, Genie 3 has some rough spots. Google admits generated worlds might stray from prompts, ignore real-world physics, and make characters less reactive. The Verge noted the tool blocks certain prompts involving licensed characters and has tightened rules on third-party content. Rivas explained that Genie 3 was “trained primarily on publicly available data from the web,” and Google is carefully monitoring user feedback as it expands access. 4
Google plans to extend Project Genie beyond U.S. Ultra subscribers eventually but didn’t provide a timeline. For now, the rollout is a controlled trial with short sessions, strict limits, and close tracking of what users actually build.
