Google DeepMind has unveiled Genie 4, an AI model capable of generating interactive virtual environments featuring realistic physics, audio, and character dialogue. The duration of user interaction within these generated worlds has been extended to two minutes, doubling the capacity of the previous version.
Users can modify the virtual environment in real time using text commands to adjust weather, add objects, or change camera angles. The system supports navigation via standard keyboard controls. The model now provides improved simulation of physical interactions, including collisions and aerodynamics, while generating ambient sounds and dynamic speech.
The Gemini 3.8 Flash model manages the consistency of the generated scenes by breaking down tasks and monitoring execution. Google also showcased the integration of Genie 4 with Gemini 4 Argon, demonstrating a virtual humanoid robot named Apollo playing checkers.
The company intends to apply this technology to film pre-visualization, game prototyping, and robotics training simulations. Genie 4 is currently limited to a select group of testers, with future plans to offer access to AI Ultra subscribers and developers via the Gemini API and Google AI Studio.
