Google DeepMind Adds Native Audio to Its Genie 4 World Model

Google DeepMind announced Genie 4 on Oct. 27 as a real-time world model that creates interactive environments with ambient sound and dialogue. Users can move through scenes with keyboard controls and change them with text prompts. Dialogue can respond to user input or events in a scene.

“Genie 4 is our first real-time world model to generate audio natively,” Google DeepMind said. Gemini 3.8 Flash can observe a simulation, break a goal into steps and check progress.

World models simulate aspects of an environment so an agent can predict how it may change and how its actions may affect it. Genie 4 lets users steer those generated environments as they unfold, combining interactive scenes with audio.

Genie 3, announced in August 2025, generated interactive environments at 24 frames per second and 720p, with scenes remaining consistent for a few minutes. Genie 4 adds native audio to the interactive environment format.

Google also demonstrated Genie 4 and Gemini 4 Argon simulating Apptronik’s Apollo humanoid playing checkers. No real-world robot results were reported. The demonstration showed a simulated task, not a reported transfer of performance to a physical robot.

Testers in creative fields, journalism, education, gaming and engineering are exploring uses such as scene previews, interactive prototypes and scenario generation. Potential applications also include training environments for robots and AI agents, though no independent performance results for those tasks were provided.

The research preview is available to trusted testers. Google plans to offer access soon to Google AI Ultra subscribers through Project Genie and to early-access partners through the Gemini API and Google AI Studio. No rollout date was given.

Generated video and audio will include SynthID. Google says scenes may lose visual detail or track of objects. It is still studying complex physics, audio and spatial accuracy, readable text and consistency over longer sessions.