Google DeepMind Releases Genie 4 World Model With Real-Time Audio and Editable Scenes

According to TechCrunch, users can type prompts in the middle of a simulation to change the scene, and the model generates audio that matches the simulated environment, including music, ambient noise and interactive dialogue. The report was cited by the technology site Slashdot.

Genie 4 follows Genie 3, which Google released as a research preview a little over a year ago. Beyond the higher level of photorealism and the addition of sound, the new model allows freeform edits such as changing the weather or the camera angle in real time, or adding new people and objects to a scene. It can hold a scene consistent for up to two minutes, double the limit of its predecessor, and the company says it offers improved physics and more human-like characters.

Tejas Kulkarni, a research director at Google, described the direction of the work during a briefing. "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," he said.

Genie 4 is available now to a limited group of testers. Google plans a broader rollout in the coming weeks for subscribers to Google AI Ultra, its top-tier consumer AI plan. A demonstration of the model has been posted on YouTube.

Google has positioned world models as a research area distinct from text and image generation, with simulations intended to respond to user input continuously rather than produce a single fixed output. The jump from Genie 3 to Genie 4, roughly a year apart, brings audio and real-time editing into the same preview stage that previously offered visuals alone.