Google has unveiled Genie 4, a new AI world-simulation model designed to generate interactive virtual environments that users can explore and modify through natural-language prompts. The company says the system is its most realistic version so far, with generated worlds able to respond as users alter the scene in real time.
Users can request changes such as different weather, a new camera position or additional characters without using conventional scene-building tools. Genie 4 also produces audio for its simulations, including ambient sound, music and interactive dialogue between characters.
Google has integrated Gemini 3.8 Flash as an agent that continuously monitors the simulation, analyzes the generated visuals and breaks complex objectives into smaller steps. The agent checks whether actions have been completed, then adjusts its plan when a user intervenes or the environment changes.
According to Google, that approach can model complex events involving cause-and-effect relationships and procedural behavior. The company also positions Genie 4 as a way to test AI-system behavior across simulated scenarios before deploying those systems in the real world.
Genie 4 is currently available only to a limited group of testers. Google plans to broaden access through Project Genie for Google AI Ultra subscribers and to early-access partners using the Gemini API and AI Studio, but it has not announced a date for wider availability.
Google DeepMind has published demonstrations of the model, while the timing and scope of its broader rollout remain unconfirmed.
