Google DeepMind Unveils Genie 4, an AI World Model With Sound and Live Editing

Google DeepMind on Tuesday, Oct. 27, 2026, unveiled Genie 4, a new AI world model that builds interactive, photorealistic scenes you can steer with plain-language prompts while they run, and that generates its own sound, from wind and music to spoken dialogue. The model is out now only as a research preview for a small group of trusted testers, with a wider rollout promised in the coming weeks.

World models are a different kind of AI from chatbots or video generators. Instead of producing a fixed clip, they simulate a place that responds as you move through it or change it. DeepMind framed the release as part of its long-term research push, writing in its announcement that understanding and simulating the world “is a crucial aspect of building AGI.”

What’s new in Genie 4

According to the DeepMind blog post, written by Raia Hadsell, Diego Rivas and Hang Qi, Genie 4 is “natively multimodal,” meaning audio is generated at the same time as the visuals rather than added afterward. That covers ambient noise and music, and also interactive dialogue from people inside a scene. Ars Technica noted that this is the first DeepMind world model with native audio, and that realistic speech could be the biggest leap if it works as promised.

The other headline feature is real-time editing. Users can type instructions mid-simulation to change the weather, move the camera, or drop in a new person or object, and Google says Gemini adapts the request to fit what’s already happening in the scene. Between prompts, Gemini also monitors and steers the simulation so it keeps moving forward in a coherent way.

Scenes can now hold together for about two minutes, double the roughly one-minute limit of Genie 3, TechCrunch and The Verge reported. Google also says Genie 4 has better physics and more human-like characters than its predecessor, which arrived as a research preview a little more than a year ago.

How Gemini powers the simulations

Google has tied Genie 4 closely to its Gemini family of models. DeepMind says Gemini Omni anchors the simulations in real-world appearance and behavior, while other Gemini models supply the reasoning needed to keep a scene consistent. The Verge reported that the system uses Gemini 3.8 Flash under the hood.

TechCrunch described Gemini as acting like a director that plans steps and checks frames while Genie creates the visuals. In one research experiment tied to Gemini Robotics, a Gemini model broke a task into steps that Genie 4 then simulated in real time, a possible way to rehearse robot scenarios before trying them in the physical world.

“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 press briefing, according to TechCrunch.

Who can use Genie 4, and when

For now, only trusted testers have access. Google says the model is “coming soon” to Google AI Ultra subscribers through Project Genie, its web tool for generating interactive worlds from text or image prompts, and to early-access partners of the Gemini API and Google AI Studio. A waitlist is open. Ars Technica reported that developer pricing hasn’t been announced and that the consumer rollout will start with Google’s most expensive AI Ultra plans.

Google is pitching uses that include training environments for robots and AI agents, prototyping game ideas and pre-visualization for filmmakers. Rivas told reporters that the real-time element is most useful in early creative work such as ideation and storyboarding, “anytime you’re kind of in need of rapid ideation.”

Limits Google acknowledges

DeepMind was unusually direct about what the model can’t do yet. Its post says visual details and object permanence can shift or drift as a simulation runs longer, after several complex edits, or as a user moves around a scene. Complex physical interactions, accurate real-time audio, legible text inside the world, spatial accuracy and long-term consistency all remain open research problems. Rivas stressed in the briefing that it is still early and that the final product experience isn’t settled.

The launch lands in a busy and consolidating market. TechCrunch noted that AI video startup Runway released its second world model a couple of months ago, that Fei-Fei Li’s World Labs is being acquired by AMD, that Amazon is in talks to buy Decart, and that Rivian acquired Third Dimension to bring simulation in-house for self-driving tests. Genie 4 is Google’s bid to stay ahead in that race.

For more on how the biggest AI labs are positioning their newest systems, see our reports on Anthropic’s Claude Duet agent preview and OpenAI’s plan to train frontier models with no internet access, plus Bill Gates’ warning about two decades of AI disruption.

Sources

• Google DeepMind: Genie 4, a new generation of world simulation
• TechCrunch: Google’s Genie 4 world model adds audio, more realistic worlds, and real-time steering
• The Verge: Google’s next Genie AI model generates more realistic experiences
• Ars Technica: DeepMind unveils Genie 4 world model