Google has opened access to Project Genie — an experimental service for creating interactive virtual worlds. This product started to become available in phases starting January 29 for Google AI Ultra subscribers in the U.S.
Project Genie is based on the new Genie 3 model, developed by Google DeepMind, which is positioned as a “universal world model.” It can generate environments, physics, and interactions in real-time depending on user actions. This makes the service unique because, unlike static 3D scenes, the behavior in worlds created by this prototype is varied and dynamic.
Google emphasizes that such models are important for the development of artificial general intelligence (AGI) because they can simulate a multitude of real-world scenarios, from robotics to animations and historical reconstructions. Project Genie is available as a web application, which combines several AI models, including Genie 3, Gemini, and Nano Banana Pro.
The main features of the service are the creation and remixing of worlds, real-time space exploration, and the customization of different vantage points. Users can describe the world using text, use images to create environments, as well as edit existing worlds and capture walkthrough videos.
It’s worth noting that this is a research prototype, so it has certain limitations. For instance, generated worlds may not always match the prompts, the duration of generation is limited to 60 seconds, and some features of Genie 3 are not yet available. The company plans to gradually expand the geographic availability of Project Genie.
According to information from Google, the company is working on servicing larger areas and plans to eventually open the technology to a wider audience.
| Features of Project Genie | Description |
|---|---|
| World Creation | Description of the world through text and images, creation of unique environments and characters. |
| World Exploration | Interact with the environment in real-time, ability to change the camera angle. |
| World Remixing | Editing existing worlds, developing variations based on others’ ideas. |
| Limitations | Generation time up to 60 seconds, possible inaccuracies in prompt matching. |
