Inside these worlds, characters are in motion. They can walk, fly or drive through landscapes that continue to unfold as the user moves. Unlike static 3D scenes, Genie generates new terrain in real time. The system also remembers what it has already created, meaning previously visited areas stay visually consistent when users return.

Because of the heavy computing power required, each session is currently limited to 60 seconds. During that time, the user is assigned a dedicated processing “chip” to power their private experience. Access to Project Genie is rolling out to subscribers of Google’s AI Ultra plan, which costs $250 a month and is currently limited to users in the US who are 18 or older. Google says wider access is planned in the future.

Behind Project Genie is a bigger ambition. Google describes Genie 3 as a “world model”, an AI system built to simulate how environments behave and respond to actions. Rather than focusing on narrow tasks like board games, these models aim to replicate the unpredictability and variety of the real world. This kind of technology could one day support everything from robotics training and architecture design to games, filmmaking and historical simulations.

Project Genie also allows users to remix existing worlds by building on prompts created by others. A gallery of curated experiences and randomised worlds offers inspiration, and users can download videos of their explorations to share later.

Google is careful to call Project Genie an experimental prototype. The company admits that the worlds may not always look realistic, physics can behave oddly, and character controls can feel laggy. Some advanced features discussed last year have not yet been included.

Still, as rivals like Runway, World Labs and other AI researchers push toward similar goals, Project Genie signals that AI-generated worlds are no longer just a concept. They are becoming something users can step into even if only for a minute at a time.