Project Genie offers three core features: World Sketching (world creation using Nano Banana Pro and Gemini), World Exploration (navigation), and World Remixing (modifying existing worlds).

Google notes the prototype still has limitations: worlds don’t always look realistic, character responses can lag, and generations are capped at 60 seconds. Some features teased in August—such as promptable events—are not yet included. An expansion to additional countries is planned.

The broader idea behind these world models is that, once mature, they could serve as flexible training environments for AI agents—allowing them to learn from simulated experience rather than relying only on static data training.