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

Google DeepMind announced Genie 4 on October 27, the next generation of its world model for real-time simulation. The model generates audio alongside video, and users can change the scene mid-simulation with text prompts.

Google presents Genie 4 as a model that brings, in its words, "unprecedented realism and native audio." For now it is an announcement backed by demos on the DeepMind blog, not a product available to everyone.

What Genie 4 can do

The model builds on the Gemini family. According to DeepMind, Gemini Omni anchors simulations in real-world appearance and behavior, while the other Gemini models handle the reasoning that keeps a simulation coherent and continuous.

Users can set an event in motion and steer it in natural language as it unfolds: move the camera, change the weather or add a character. Every world has its own soundtrack, covering ambient sound, music and dialogue that reacts to what the user does.

TechCrunch adds that physics and character realism have been improved, and that a scene stays consistent for up to two minutes, twice as long as in Genie 3. According to the outlet, Gemini acts as the director, planning steps and checking frames, while Genie renders the visuals.

Robots and agents in simulation

DeepMind says the simulations are meant for safely training robots and agents, and for creative tools in filmmaking, journalism, gaming and manufacturing. The company explicitly describes simulation as a crucial step toward AGI.

TechCrunch describes an experiment tied to Gemini Robotics: a Gemini 4 Argon model broke tasks down into steps, and Genie 4 simulated them in real time. According to the outlet, the experiment shows a possible way to test robot scenarios in simulation before they happen in the real 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, research director at Google (quoted by TechCrunch)

Limitations

The team acknowledges that details and objects can drift as a simulation runs longer. Per TechCrunch, it also has not solved the problem of audio accuracy or the legibility of text inside the world.

Diego Rivas of DeepMind stressed that it is early days and that the end-to-end product experience is not yet clear. According to Rivas, using Gemini Omni to turn prototypes into finished creative assets is still a ways off.

Competition and availability

TechCrunch notes that Genie 3 arrived as a research preview a little over a year ago, and that a couple of months ago Runway released GWM Worlds 2, with playable worlds and audio. Google did not give pricing, benchmarks or exact release dates in the announcement.

For Polish game developers and robotics teams, the key question will be whether access through the Gemini API and AI Studio actually materializes. For now it is an announcement for a closed group.