Google launches Project Genie to create virtual worlds

Access to the prototype became available on January 29th to Google AI Ultra subscribers in the US. The service is planned to expand to other territories later. Project Genie allows users to create, explore, and modify their own interactive worlds.

The project is based on the Genie 3 world model developed by Google DeepMind. It generates environments in real time as the user moves and interacts, simulating the physics and dynamics of worlds. This distinguishes it from static 3D scenes.

The prototype operates through a web application and combines three key functions:

• World Sketch: Create an environment using text queries, generated or uploaded images. The user can define the character, the world, and the mode of movement.
• World exploration: Navigate through a generated environment that expands in real time.
• World Remix: Modify existing worlds based on their original requirements. You can also explore the examples in the gallery for inspiration.

The project is positioned as exploratory, and the current model has limitations. Generated worlds may not always accurately match requirements or realistic physics, and character control and lag may be reduced. The duration of a single generation is limited to 60 seconds. Some previously announced features, such as modifiable in-world events, have not yet been implemented.

Google's goal is the responsible development of AI technologies toward artificial general intelligence (AGI). The launch of the prototype to a limited audience is aimed at exploring the application of global models in various fields, including robotics, animation, and gaming.