Google DeepMind Launches Genie 4 Real Time Interactive World Model With Native Audio, Human Dialogue, and Gemini Reasoning
This is Genie 4, our most realistic and interactive world model yet. 🌐 Explore dynamic environments in real time to see how the scene responds, sounds, and how the people in it move, react, and speak.
Context
Combining two minute coherent 3D world generation, native environmental audio, interactive human speech, and live Gemini 3.8 Flash supervision turns world models from silent physics demos into closed loop simulation arenas for training embodied robots and software agents.
Google DeepMind introduced Genie 3 as a research preview in 2025, demonstrating interactive 3D world generation from prompts with up to one minute of visual scene consistency but without native audio or conversational human characters. World simulation has since become a primary battleground for frontier labs seeking synthetic training environments for robotics and autonomous agents.
Deep dive
On October 27, Google DeepMind unveiled Genie 4, its newest real time interactive world model authored by Raia Hadsell, Diego Rivas, and Hang Qi. Built on capabilities from the Gemini family, Genie 4 uses Gemini Omni to anchor virtual scenes in lifelike physical appearance and human behavior while pairing with reasoning models to maintain scene coherence for up to two minutes, doubling the duration limit of Genie 3.
Users navigate generated worlds in real time using standard keyboard and camera controls while typing natural language prompts mid simulation to alter weather, move the camera, or introduce new characters and objects on the fly. Genie 4 advances physical simulation across aerodynamics and object collisions, and delivers a major leap in simulating human characters that move naturally and react organically to surrounding events.
Crucially, Genie 4 is Google DeepMind's first real time world model to generate audio natively alongside video frames. Every simulated world produces matching environmental soundscapes such as wind or crashing waves, background music, and adaptive human dialogue where simulated characters speak and respond to user inputs in real time. To simulate complex multi step events, Google DeepMind integrates Gemini 3.8 Flash as a guiding agent in the loop that continuously observes incoming frames, decomposes high level goals into sequential steps, and verifies whether each action succeeded. In a parallel robotics experiment with Gemini 4 Argon, researchers used Genie 4 to test multi step robot task execution in simulation before real world deployment.
What to expect next
Google DeepMind is initially rolling out Genie 4 as a research preview to trusted testers before expanding access in the coming weeks to Google AI Ultra subscribers through Project Genie and early access partners in the Gemini API and Google AI Studio. Watch how quickly robotics teams adopt the combined Gemini 3.8 Flash and Genie 4 loop to stress test embodied policies before physical deployment.