Google unveils Genie 4, a world model with audio and live steering

Google DeepMind released Genie 4, its newest world model, in research preview on Tuesday. The system generates interactive, photorealistic virtual worlds and lets users type prompts mid-simulation to change what’s happening on screen, according to the company.

Genie 4 can hold a scene together for up to two minutes before details start to drift, Google says, twice as long as its predecessor Genie 3, which launched a little over a year ago. The new model also adds real-time audio that matches the simulated world, including music, ambient noise, and dialogue that responds to what’s happening in the scene. Users can make free-form edits on the fly, such as changing the weather, shifting the camera angle, or adding new people or objects, and Google says the model now offers better physics and more humanlike characters than Genie 3.

The release ties Genie more closely to Gemini, Google’s family of AI models. Gemini now acts as a kind of director, planning steps and checking frames while Genie generates the visuals. In one research experiment tied to Gemini Robotics, a Gemini 4 Argon model broke a task into steps that Genie 4 simulated in real time, a possible way to test robot behavior in simulation before trying it 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 briefing, according to TechCrunch.

Google is pitching gaming, interactive learning, and robotics as commercial use cases, including a speculative example of stepping into a historical painting and talking to its subjects. But Diego Rivas, a Google DeepMind engineer, told TechCrunch that early testers are also pointing to film prototyping as a promising use. “The real time aspect becomes most valuable in the areas where keeping a state of flow in the creativity space becomes incredibly useful,” Rivas said, citing ideation and storyboarding as examples. He added that filmmakers could eventually use Gemini Omni, a separate multimodal model, to turn Genie-generated prototypes into finished creative assets, though he said that capability is still a ways off.

Still early, Google says

Throughout the briefing, Rivas stressed that it’s still early days for Genie 4 and that the eventual product experience isn’t settled. The team also acknowledged gaps: details and objects can drift the longer a simulation runs, and Google hasn’t yet figured out how to keep audio accurate or make in-world text legible.

The release lands in a world-model field that’s consolidating fast. AMD is acquiring Fei-Fei Li’s World Labs, Amazon is reportedly in talks to buy Decart, and Rivian just bought Third Dimension to build simulation capability for self-driving tests in-house. Runway, meanwhile, released its own playable world model with audio, GWM Worlds 2, just a couple of months ago.

Genie 4 is available now only to Google’s trusted testers. Over the coming weeks, it’s set to reach Google AI Ultra subscribers through Project Genie, along with early access partners on the Gemini API and Google AI Studio. Whether Genie 4 ships as a standalone product, a feature inside existing tools, or something else entirely remains an open question Google has not answered.