Google DeepMind on Tuesday launched Genie 4, its newest world model, in research preview for “trusted testers.” Billed as an AI world simulator, the release delivers three headline upgrades: dramatically more photorealistic worlds, real-time audio generation for the first time, and — most striking — real-time steering. Users can type natural-language prompts mid-simulation to change the weather, shift the camera angle, or add new people and objects, and the world responds instantly.
Two-Minute Consistent Worlds, Now With Sound
Compared with Genie 3, released as a research preview a little over a year ago, Genie 4 maintains a consistent scene for up to two minutes — double its predecessor’s limit — while delivering better physics and more humanlike characters. On the audio side, the model generates music, ambient noise, and interactive dialogue that match the simulated world, making the experience fully “sound and vision” for the first time.
Gemini as the Director: A Hybrid World Model Architecture
The other key change is deep integration with Google’s flagship AI model. In the new architecture, Gemini 3.8 Flash acts as the “director” — planning steps and checking frames — while Genie 4 creates the visuals. “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,” said Tejas Kulkarni, research director at Google.
In a research experiment tied to Gemini Robotics, a Gemini 4 Argon model broke tasks into steps that Genie 4 then simulated in real time — a demonstration that robot scenarios could be tested in simulation before they happen in the physical world, with major implications for training cost and safety.
The World Model War Heats Up
Genie 4 arrives as the world model industry consolidates. AI video startup Runway released its second world model, GWM Worlds 2, two months ago — a playable model that responds dynamically to user input and includes audio generation. Fei-Fei Li’s World Labs is being acquired by AMD, Amazon is in talks to buy Decart, and Rivian just acquired Third Dimension to bring simulation in-house for self-driving tests. Every major player has reached the same conclusion: real-time generative, interactive simulation is infrastructure for the next generation of AI applications.
Google’s edge is holding both halves of the stack — Genie’s generative power and Gemini’s reasoning and planning — a “generator plus director” combination no rival currently matches end to end.
From Games to Film Pre-Production: The Surprise Use Case
Officially pitched use cases include gaming, interactive learning (imagine stepping into a painting of a historical scene and chatting with its subjects), and robotics training. But the earliest adopters among trusted testers have surfaced a different favorite: film prototyping. “The real-time aspect becomes most valuable in the areas where keeping a state of flow in the creativity space becomes incredibly useful — for example, in the ideation phase, in the storyboarding phase,” said Diego Rivas, a Google DeepMind engineer, during a press briefing.
Rivas added that users could eventually rely on Gemini Omni, a multimodal model that generates video from images, audio, and text, to turn those prototypes into creative assets — though those capabilities remain some ways off. Google also acknowledges Genie 4’s gaps: details and objects drift the longer a simulation runs, and the team hasn’t fully solved audio accuracy or legible in-world text.
Availability and What to Watch
Genie 4 is available now as a research preview for trusted testers. In the coming weeks it rolls out to Google AI Ultra subscribers in Project Genie — Google’s tool for generating interactive worlds from text or image prompts — plus early access partners of the Gemini API and Google AI Studio. A public waitlist is open.
The bigger picture: world models are converging with robotics, autonomous driving, and game engines at speed. Once simulated worlds become cheap and realistic enough, “try ten thousand times in virtual, then once in reality” becomes standard engineering practice — and that is the ultimate prize Google, AMD, and Amazon are all racing for.
Conclusion
Genie 4 matters for more than prettier pixels. Real-time steering, real-time audio, and a Gemini director together turn the world model from a “player” into an “instrument.” For creators, it’s an ideation superpower; for engineers, a cheap sandbox for robots and self-driving. The world model war has only begun — but the fight for its command layer is already white-hot.
Sources: TechCrunch, The Verge, Ars Technica (October 27, 2026 reporting)
