The most interesting thing about Genie 4 is not the graphics. It’s that you can type a prompt in the middle of a simulation and the world changes while you’re standing in it. Rain starts, the camera moves, a new person walks into frame. That’s a different product from a model that renders a pretty scene and then ignores you.
Google DeepMind released Genie 4 in research preview on Tuesday. Compared with Genie 3, which came out a little over a year ago, it is far more photorealistic, has better physics, and renders more human-like characters. It also generates real-time audio that matches the scene, including music, ambient noise, and interactive dialogue. Scene consistency doubles to two minutes. That is progress, but two minutes is still a demo, not a game level.
The Gemini integration may matter more than any of that. Gemini acts as a director, planning steps and checking frames, while Genie produces the visuals. In a Gemini Robotics experiment, a Gemini 4 Argon model split tasks into steps and Genie 4 simulated them live. The idea is to test robot behavior in simulation before it happens in the real world. Google research director Tejas Kulkarni said we will see more of these hybrid models, with Gemini writing code and text while Genie simulates reality at different levels of abstraction.
The competitive context explains the urgency. Runway released its second world model, GWM Worlds 2, a couple of months ago, and it is also playable and also generates audio. Meanwhile the category is consolidating fast. AMD is acquiring Fei-Fei Li’s World Labs, Amazon is reportedly in talks to buy Decart, and Rivian bought Third Dimension to bring simulation in-house for self-driving tests. When chipmakers, cloud giants, and carmakers start buying world model startups, the technology is being treated as infrastructure, not a research curiosity.
Google’s pitched use cases are the familiar ones: gaming, interactive learning, and robotics. One speculative example is stepping into a historical painting and chatting with its subjects. The more grounded idea is coming from trusted testers, who are using Genie 4 to prototype films. DeepMind engineer Diego Rivas said real-time generation is most valuable for ideation and storyboarding, where keeping a state of flow matters. That puts Genie next to Veo, Google’s video model with synchronized audio, which filmmakers already use. Rivas also floated Gemini Omni as a way to turn prototypes into finished assets, though he said that is a ways off.
The limits are real, and Google admits them. Details and objects drift the longer a simulation runs. The team has not figured out how to make the audio reliably accurate or in-world text legible. Rivas also said it’s unclear what the end-to-end product will look like. So this is a research preview in the truest sense.
Availability is narrow for now. Trusted testers have access today. In the coming weeks, Google AI Ultra subscribers will get it in Project Genie, and early access partners will get it through the Gemini API and Google AI Studio.
Does it matter? Yes, mostly for the robotics angle and the Gemini pairing. Runway still has a head start on creative tooling, and no one has solved long-horizon consistency. But Google is the only player combining a frontier language model, a video model, and a world model under one roof. If drift gets fixed, that combination will be hard to match.
