Google Genie 2 Howto

Note: Genie 2 is not publicly available: Based on the sources, Genie 2 is a research model by Google DeepMind that is not currently released for public use. It is being used internally for AI research and development.

Input an image prompt: If you had access, you would start by providing a single image prompt (either generated by Imagen 3 or a real photo) to define the virtual environment you want to create.

Wait for environment generation: Genie 2 would process the image prompt and generate an interactive 3D environment based on it. This environment can last up to 60 seconds, with most examples lasting 10-20 seconds.

Control with keyboard/mouse: Once the environment is generated, you can control movement and interactions using standard keyboard and mouse inputs. The model recognizes which elements should be controllable (like characters) versus static elements (like trees).

Explore the environment: You can move around, interact with objects, and explore the generated world. The model maintains consistency and remembers areas even when they're not in view.

Optional: Deploy AI agents: For research purposes, AI agents like SIMA can be deployed to interact with and navigate the generated environments following natural language instructions.