Google Launches Project Genie, an Interactive World Generator Built on Genie 3

Google DeepMind released Project Genie on January 29, 2026, a research prototype that turns a text description or a photo into an explorable, interactive world generated in real time. The tool is currently available only to adult subscribers of the Google AI Ultra plan in the United States, though the company says it plans to expand to more countries.

Project Genie rests on three pillars Google calls world sketching, world exploration, and world remixing. The first lets users sketch a world with a text prompt or an uploaded photo, with a preview generated by Nano Banana Pro before they even step into the environment. Users choose a perspective, first-person or third-person, and a way of moving through the world: walking, driving, or flying.

How on-the-fly generation works

The second pillar, world exploration, handles navigation itself. Instead of rendering a pre-built, static 3D model, Genie 3 generates the next stretch of the world on the fly, depending on where the user goes and what actions they take. Google describes this as generating the path ahead of the user in real time, based on the decisions they make, rather than pulling from a pre-planned map.

The third element, world remixing, lets users turn existing worlds into new versions. Google has put together a curated gallery of sample worlds along with a randomizer, and generated exploration sessions can be saved as video and downloaded.

Limitations and training scale

Google's official announcement does not give a single, clearly stated number of training hours for the Genie 3-based model, but the company is upfront about the tool's limitations. Generated worlds may not look fully realistic and don't always stick closely to the prompt, the uploaded image, or the laws of physics. Characters in the environment can be harder to control and respond with noticeable lag, and every exploration session is capped at 60 seconds of generation. Also missing is the promptable events feature, the ability to change events mid-exploration, which Google had promised back in August 2025 when Genie 3 itself launched.

Genie 3 generates the path ahead of you in real time based on the actions you take. - from Google DeepMind's announcement

Uses beyond entertainment

Google frames Project Genie as a first step toward bringing to a wider audience a technology that had previously reached only select researchers and testing partners. The announcement was signed by Diego Rivas, group product manager at Google DeepMind, Elliott Breece, product manager at Google Labs, and Suz Chambers, director of Google Creative Lab. In the post, they point to uses that go beyond entertainment, from robotics simulation to animation and fiction prototyping to recreating real-world locations and historical scenarios.

The tool isn't yet available for users in Poland, since Google is limiting the launch to the US market and the Ultra plan, the company's most expensive subscription tier. It's a recurring pattern with Google's newest, most computationally expensive features, some Gemini video features rolled out the same way before. The company hasn't specified a timeline for expanding to other countries or cheaper plans.

Project Genie fits into the broader trend of world models, AI systems that don't just generate images or video but simulate interaction and the consequences of actions in real time. Google DeepMind and its rivals, including several Chinese labs, have spent months showcasing similar prototypes as a possible direction for games, training simulations, and synthetic training data for robots. Project Genie is the first version of this technology released to paying users rather than just shown in demo material.