Google DeepMind released Genie 4 in research preview on Tuesday, 27 October, and the headline features are the obvious ones: photorealistic interactive worlds, real-time audio that matches the scene (music, ambient noise, interactive dialogue), and free-form edits — change the weather, swing the camera, add a person or an object — typed mid-simulation. The scene stays consistent for up to two minutes, double the limit of Genie 3, which shipped a little over a year ago, with better physics and more humanlike characters.
The interesting release is not the pixels. It is who is directing them.
Gemini Writes, Genie Renders
In a research experiment tied to Gemini Robotics, a Gemini 4 Argon model broke a task into steps and Genie 4 simulated those steps in real time. Gemini plans the sequence and checks the frames; Genie produces 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, during a briefing.
That reframes the world model from a toy into a rehearsal stage. If a task plan can be rendered before a robot touches hardware, failures become cheap. It also explains why the field is consolidating: Runway shipped its second playable world model, GWM Worlds 2; AMD is acquiring World Labs for $8.2B; Amazon is in talks to buy Decart for around $7B; and Rivian bought Third Dimension on 22 October to bring simulation in-house for self-driving testing.
Where the Limits Are
Google was candid about the gaps. Details and objects drift the longer a simulation runs, in-world text is not reliably legible, and the accuracy of the generated audio is still an open problem. The end-to-end product shape is undefined — the team called it “still early days.”
Commercially, the pitch is gaming, interactive learning (chat with the subjects of a painting) and robotics. Trusted testers are using it for something else: film prototyping, where staying in flow during ideation and storyboarding is the whole value. Diego Rivas, a Google DeepMind engineer, noted that Gemini Omni could eventually turn those prototypes into assets — though that capability “is still a ways off.”
Genie 4 is live for trusted testers now, rolling out over coming weeks to Google AI Ultra subscribers in Project Genie and to early access partners of the Gemini API and Google AI Studio.
Related: TechCrunch on Genie 4 and AMD’s World Labs acquisition