Google DeepMind unveils Genie 4: a world model that talks back, with native audio and live steering

Google DeepMind has a new world model, and this one talks back. Announced on October 27, 2026, Genie 4 is the lab's latest system for real-time interactive simulation — a model that generates a world you can walk through, steer with typed prompts, and now hear. It is the first real-time world model from DeepMind to generate audio natively, so a beach scene arrives with crashing waves and its characters can hold a conversation that adapts as the scene changes.

The announcement, authored by Raia Hadsell, Diego Rivas and Hang Qi, frames Genie 4 as a step toward a much older goal: an AI that understands the world well enough to simulate it. DeepMind says the model "unlocks unprecedented realism and improves real-time interactive simulation," with enhanced photorealism, fluid movement, and native audio — every sound in its demos produced by the model itself.

Sound is the headline#

Earlier world models were silent films. Genie 4 is the first in the series to carry its own soundtrack: every world generates ambient sound for its setting, from wind in the trees to waves on a beach, plus music and interactive dialogue. Characters converse and react to what you type or to events unfolding in the world, in real time.

That matters because sound is half of what makes a simulated place feel like a place. A silent world, however photorealistic, reads as a render. A world that answers when spoken to reads as somewhere you are standing — which is exactly the illusion DeepMind is chasing, and the reason the lab calls simulated people its most dramatic leap in this release: characters that move naturally, respond to their surroundings, and can be steered interactively.

Steering a world as it runs#

The second change is control. With Genie 4, you can rewrite a running simulation in plain language — change the weather, move the camera, introduce a new person or object — and Gemini, working behind the scenes, adapts your prompt to fit the scene already on screen. Between your inputs, Gemini keeps monitoring and steering the world so it progresses coherently instead of drifting.

Under the hood, two Gemini roles do the work. Gemini Omni anchors the simulations in real-world appearance and behavior, while other Gemini models supply the reasoning that keeps a scene coherent over time. For multi-step events, Gemini 3.8 Flash acts as a guiding agent in the loop: it watches the scene, breaks a goal into a sequence of steps, reviews incoming frames, and verifies each step succeeded — re-planning on the fly when the world shifts or the user intervenes.

A rehearsal stage for robots — and filmmakers#

DeepMind is pitching Genie 4 well beyond entertainment. The same machinery that lets a user steer a scene lets researchers test how AI systems navigate environments before anything touches the real world. In one experiment, Gemini 4 Argon and Genie 4 together simulate Apptronik's Apollo humanoid playing checkers, with Argon breaking the task into steps that Genie renders in real time — a path toward simulating a breadth of scenarios in imagined worlds.

Trusted testers across creative industries, journalism, education, gaming, and engineering are already trying other uses. Filmmakers are testing Genie 4 as a rapid "draft mode" for pre-visualization, running multiple takes of a scene live and adjusting camera angles, lighting, dialogue, and action on the fly. Photojournalists are exploring how a single archival photograph can become a grounded simulation of an environment in seconds, including visualizations of short-term events like a flash flood or wildfire. Game developers are going from an idea to an explorable prototype in seconds, testing visual styles and mechanics through prompts before writing engine code.

What it isn't yet#

DeepMind is unusually candid about the limits. Genie 4 remains an early, experimental model: visual details and object permanence can still shift or drift over time, especially as a simulation runs longer, after multiple complex prompt edits, or while navigating through a scene. Complex physical interactions, real-time audio accuracy, rendering legible text, spatial accuracy, and long-term consistency are all listed as active research areas.

The rollout reflects that caution. Genie 4 is available now only as a research preview to trusted testers, after internal safety evaluations aimed at preventing harmful or malicious content. It is coming soon to Google AI Ultra subscribers in Project Genie and to early access partners of the Gemini API and Google AI Studio, with a waitlist open for Project Genie. And to keep the output identifiable, every video and audio clip Genie 4 produces will carry SynthID, Google's imperceptible watermarking technology, across both visual and audio modalities.

None of that dims the direction of travel. DeepMind's next step, it says, is hybrid world models — Gemini mapping the structure and narrative timeline behind the scenes while Genie renders the immediate physical reality, pixel by pixel. If that combination holds together, the simulated worlds stop being demos and start being rehearsal stages: places where robots, filmmakers, and researchers can try things out before the real world has to bear the cost of being wrong.

Sources: Google DeepMind's announcement, "Genie 4: A new generation of world simulation" (October 27, 2026); Google's announcement, "We're announcing Genie 4, our new world model for real-time interactive simulation" (October 27, 2026); Unite.AI's coverage of the research preview rollout. The editorial illustrations are AI-generated.