Google DeepMind has announced Genie 4, a new world model that generates explorable, photorealistic interactive worlds in real time from a text prompt or an image. Announced on October 27, 2026, it is available as a research preview for trusted testers, rolling out first 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 others.
The headline change is sound. Genie 4 is DeepMind's first real-time world model with native audio: music, ambient noises, and interactive dialogue are generated simultaneously with the world itself. Characters in the scene can converse and react in real time. That turns a world from something you watch into something you can talk to.
The second change is duration. Coherent scenes now last up to about two minutes, double the 60-second limit of Genie 3. Two minutes is still short, but it is long enough for a scene to develop: a character can cross a room, weather can turn, a conversation can go somewhere. The jump matters because the failure mode of world models has always been drift — the longer a simulation runs, the more details slip. DeepMind is candid that this is still a limit: visual details and object permanence can shift or drift over time, especially in longer runs, after multiple complex prompt edits, or while navigating.
The third change is steering. You can now change the world mid-simulation with natural-language prompts: change the weather, move the camera, add a person or an object, while the scene keeps running. You navigate with keyboard controls, and Gemini adapts your prompts to fit the scene. Under the hood, Gemini Omni anchors the real-world appearance and behavior of objects, and Gemini 3.8 Flash acts as the in-loop guiding agent that keeps the simulation on track.
DeepMind calls people and physics the biggest leap in this version. Characters move more naturally, and the model handles aerodynamics and collisions more convincingly. That is the difference between a backdrop and a place: things in the world now behave as if they belong there, at least for a couple of minutes at a time.
All Genie 4 video and audio carries SynthID watermarking, so generated output is marked as AI-made. That is a small detail with large implications as these tools get closer to photorealism: provenance has to be built in, not bolted on later.
Why this matters beyond the demo
It is easy to file world models under "impressive demo" and move on. Genie 4 points somewhere more practical. In game design, a director could sketch a playable scene from a sentence and iterate on it live instead of waiting on a build. In film production, a team could block out a location, test camera moves, and rehearse a sequence before a set exists. In education, a teacher could drop students into a historical street or a physics experiment and let them walk around in it. These are directions the technology points toward, not commitments Google has announced — but the shape of the tool makes them easy to imagine.
Robot training is the one application with a concrete research link. DeepMind paired Genie 4 with a Gemini 4 Argon model to break tasks into steps simulated in real time, a possible route to testing robot scenarios in simulation before touching hardware. If that path works, the cost of trying a robot behavior drops from a lab session to a prompt.
The honest read is that Genie 4 is a research preview with real limits: two minutes of coherence, drift under pressure, and a trusted-tester rollout. But the trajectory is clear. Worlds that look real, sound real, respond to conversation, and accept live direction are moving from research curiosity toward a prototyping medium. The builders who learn to direct these worlds now — prompt, steer, test, iterate — will be the ones ready when the coherence window stretches from minutes to much longer.
Read DeepMind's announcement (Genie 4: a new generation of world simulation) and TechCrunch's coverage (Google's Genie 4 world model adds audio, more realistic worlds, and real-time steering).
More from Terry at Airgrix.
Top comments (0)