Simulation

Genie 4: Google’s World Model Builds Explorable Worlds With a Soundtrack

co-produced by AI28. October 2026, 07:22

From prompt to explorable world, now with a soundtrack: Google DeepMind has unveiled Genie 4, the next generation of its world model. The A.I. generates interactive 3-D worlds in real time that users can move through like a video game, and for the first time it delivers matching sounds, music and dialogue along with them. Google calls it “our first real-time world model to generate audio natively.”

What Genie 4 Can Do

A world model computes an environment frame by frame and responds to keyboard or text input. Compared with its predecessor, Genie 3, which came out a little over a year ago, the new version brings several upgrades, according to DeepMind:

Sound: Genie 4 generates ambient noise, music and interactive dialogue in sync with the picture.
Live directing: Users can change a scene in the middle of a simulation by typing, for example switching the weather, moving the camera or adding people and objects.
Longer consistency: A scene stays coherent for up to two minutes, twice as long as with Genie 3, according to TechCrunch.
Physics: The model simulates aerodynamics and collisions, Google says, and characters move more naturally and react to their surroundings.

In the background, Gemini acts as a kind of director. Google’s A.I. models plan the individual steps of a task, check the generated frames and maintain continuity while Genie renders the world.

From Robots to Film Sets

Google sees several areas of use. In robotics, machines are meant to train in simulated worlds before they are deployed in the real one. DeepMind shows a demo in which an Apptronik Apollo humanoid robot plays checkers, with Gemini breaking the task into steps and Genie 4 simulating them in real time. Other uses include game prototypes, learning applications such as stepping into a historical painting and talking to its subjects, and filmmaking, where directors want to play through scenes and storyboards faster.

Genie 4 is not yet widely available. A research preview is currently limited to trusted testers. In the coming weeks, Google AI Ultra subscribers are set to get access through the Project Genie tool, as are early access partners via the Gemini API and Google AI Studio. All generated video and audio carry Google’s invisible SynthID watermark.

Where Genie 4 Still Falls Short

Google itself acknowledges limits. In longer sessions or after several edits, details can drift and objects can disappear. Complex physical interactions, accurate sound, legible text inside the world and long-term consistency remain research topics, according to DeepMind. “It’s still early days for Genie 4,” Diego Rivas of DeepMind told TechCrunch.

Skepticism comes mainly from two directions. The games industry pushed back when Project Genie launched earlier this year and shares of Unity, Roblox and Take-Two fell sharply. Industry figures considered the sell-off overblown. “You can’t prompt your way to that. Not yet. Not for a long, long time,” Andrew Green, the chief executive of the studio Storygrounds, said of complex games, according to Pocketgamer.biz. In robotics, the A.I. writer Ben Dickson has argued that a generated world has no ground truth to check against, so errors in the model could compound when robots are trained in it.

The Competition

World models are one of the hottest areas in A.I. right now. The New York start-up Runway released GWM Worlds 2, a playable world model with sound, just a few months ago. The chipmaker AMD is acquiring the world model start-up World Labs, founded by the A.I. pioneer Fei-Fei Li, for $8.2 billion (about €7.3 billion). Amazon is in talks to buy the Israeli start-up Decart, according to The Wall Street Journal, and the electric carmaker Rivian has bought Third Dimension to build simulations for autonomous driving in-house. Nvidia, too, is betting on world models for robots and self-driving cars with its Cosmos platform.