Google has released Project Genie, an experimental research prototype that allows users to create and explore interactive virtual worlds. The tool is now available to Google AI Ultra subscribers in the United States who are 18 or older.
Project Genie is powered by Genie 3, a world model that generates interactive environments in real time. Unlike static 3D snapshots, the system creates the path ahead as users move through the world. Google DeepMind developed Genie 3 to simulate physics and interactions for dynamic environments. The company states that the technology can simulate various real-world scenarios, from robotics to animation and historical settings.
The prototype offers three main capabilities. Users can create worlds through text prompts and images, defining characters and exploration methods ranging from walking to flying. The system integrates with Nano Banana Pro for previewing and modifying worlds before entering them. Once inside, users can explore navigable environments that generate content in real time based on their actions. The tool also allows users to remix existing worlds or build upon curated examples from a gallery.
Testing reveals significant limitations. Each generated world can only be explored for 60 seconds at a resolution of approximately 720p and a frame rate of about 24 frames per second. The Verge tested the prototype and reported frustrating input lag, worse than typical cloud gaming experiences. Generated worlds sometimes fail to maintain consistency—paint trails disappear, roads turn into grass, and characters occasionally become uncontrollable.
The tool successfully generated worlds resembling recognizable gaming franchises. A reporter created environments that looked like Super Mario 64, Metroid Prime, and The Legend of Zelda: Breath of the Wild. However, Google subsequently blocked the generation of Super Mario 64-based worlds due to third-party content provider interests. The system refused to generate a Kingdom Hearts-themed world when specific character names were used.
Diego Rivas, a product manager at Google DeepMind, told The Verge that the company released Project Genie to learn about new use cases. Google is interested in applications for visualizing scenes for filmmaking, creating interactive educational media, and potentially helping robots navigate the real world. Shlomi Fruchter, a Google DeepMind research director, emphasized that Project Genie is not an end-to-end product for daily use.
Google acknowledges several known issues. Generated worlds may not look completely realistic or adhere closely to prompts and real-world physics. Characters can be less controllable with higher latency. Some capabilities announced for Genie 3, such as promptable events that change the world during exploration, are not yet included in the prototype.
The Genie 3 model was trained primarily on publicly available data from the web, according to Google. This likely explains why the system can generate environments that resemble existing games—the model learned from countless videos of gameplay available online.
Google plans to expand access to more territories in the future. Users can download videos of their created worlds and explorations. The company states its goal is to make these experiences and technology accessible to more users over time.
Sources: Google Blog, The Verge
