Earlier this summer, Google DeepMind introduced Genie 3, a type of world model powered by AI that produces visuals and adapts to user movement within the simulated space it creates. DeepMind originally presented it as a resource for developing AI agents, but the company is now extending access to external individuals through the new Project Genie initiative.
Getting started requires a $250 monthly subscription to Google's AI Ultra service, along with U.S. residency and an age of at least 18. Upon release, the platform includes three interaction options: World Sketching, exploration, and remixing. World Sketching involves Google's Nano Banana Pro model producing the base image that Genie 3 employs to build the navigable environment. During this process, participants can outline their avatar, select the viewpoint—such as first-person, third-person, or isometric—and specify navigation preferences for the upcoming simulation. Prior to entering the generated scene, Nano Banana Pro displays a preliminary sketch for potential modifications. Additionally, users can apply custom descriptions to environments previously created with Genie.
It's important to note that Genie 3 does not function as a complete game engine. Although the results may resemble video games and support simulated physics, standard gameplay features are absent. Each creation is restricted to 60 seconds, and the output is limited to 24 frames per second at 720p resolution. Nevertheless, for those with an AI Ultra subscription, this represents an exciting chance to experience the latest innovations from DeepMind over recent years.