Google DeepMind's Genie 4 adds native audio and live text steering

A world model is not a video clip. You give Genie 4 a text or image prompt, and it generates a place you can move through in real time, frame by frame, using standard keyboard controls for navigation and camera. DeepMind puts it this way: you set an event in motion, and Genie 4 simulates what happens next, including how the world responds, how it sounds, and how the people in it move, react, and speak.

Gemini Omni, from Google's Gemini family, anchors the simulations in real-world appearance and behavior, DeepMind said, while other Gemini models supply the reasoning that keeps a simulation coherent. The point of the work, in DeepMind's words, is training environments for robots and agents and new interactive experiences, and the company calls understanding and simulating the world “a crucial aspect of building AGI.”

Native audio is new in Genie 4. DeepMind calls Genie 4 its first real-time world model to generate audio natively: every world carries the ambient sound of its setting, from wind in the trees to waves at a beach, and characters converse and react in real time, with dialogue that adapts on the fly.

The other change is steering. You can type a new instruction while a scene runs, changing the weather, moving the camera, or adding a person or object, and Gemini adapts the prompt to fit the scene already running. TechCrunch reports a scene can stay consistent for up to two minutes, which it describes as double Genie 3's limit. DeepMind's Genie 4 post gives no session length, and its August 2025 Genie 3 post described consistency “for a few minutes” with visual memory reaching back about one minute, so the comparison is TechCrunch's and does not match DeepMind's own figures. DeepMind says Genie 4 simulates real-world physics such as aerodynamics and collisions. DeepMind calls the simulated people the most dramatic leap: characters that move naturally, respond to their surroundings, and can be steered in real time.

Gemini 3.8 Flash works as a guiding agent inside the loop. As Genie 4 generates the world, Gemini watches the scene, breaks a goal into a sequence of steps, and reviews incoming frames to check each step succeeded, adapting its plan when the world shifts or the user intervenes, DeepMind said.

“What we’ll increasingly see in the future are these hybrid world models, where you have a combination of Gemini writing code and text, and Genie ingesting that in real time and simulating reality at different levels of abstraction,” Tejas Kulkarni, a research director at Google, said during a briefing, according to TechCrunch.

DeepMind showed the idea with Gemini 4 Argon and Gemini Robotics: Argon broke a user-instructed task into steps, fed as text captions while Genie 4 rendered, simulating Apptronik's Apollo humanoid playing checkers. The experiment points to testing robot scenarios in simulation before they happen in the real world, TechCrunch reported.

Not yet, unless you are a trusted tester. DeepMind does not name individual testers. It says it has been working with tester cohorts across creative industries, journalism, education, gaming, and engineering, who are trying Genie 4 for film pre-visualization, turning a single archival photograph into a simulation, and game prototypes. DeepMind says Genie 4 is coming soon to Google AI Ultra subscribers in Project Genie and to early-access partners of the Gemini API and Google AI Studio, and a waitlist is open. TechCrunch reports that rollout in the coming weeks. The announcement names no price and no firm date.

DeepMind lists these limits. Visual details and object permanence can shift or drift over time, especially in longer sessions, after several complex prompt edits, or while navigating. Complex physical interactions, real-time audio accuracy, legible text, spatial accuracy, and long-term consistency are still active research. All video and audio Genie 4 generates will carry SynthID, Google's imperceptible watermark. DeepMind calls Genie 4 an early, experimental model, and TechCrunch reports the team said it is not yet clear what the end-to-end product experience will be.