Google unveils Genie 4, its AI model that creates interactive virtual worlds with real-time sound

Mountain View, California — Google DeepMind has unveiled Genie 4, its new artificial intelligence model capable of creating interactive virtual worlds in real time, with natively generated sound: music, ambient noises and character dialogue, all at the same time as the image.

According to the announcement published on October 27, Genie 4 is the company's first real-time world model to generate audio inside its own process, rather than adding it afterwards. Users can explore the world with a keyboard and also modify it while it is running, using natural-language instructions: changing the weather, moving the camera or adding people and objects, without restarting the scene from scratch.

The system builds on the Gemini family: Gemini Omni anchors the appearance and behavior of objects in the real world, while Gemini 3.8 Flash acts as a guiding agent inside the simulation loop, observing each frame, breaking goals into steps and checking that each step is completed. Scenes thus stay coherent for about two minutes, double the one-minute limit of its predecessor, Genie 3, with more natural human movement and improved physics for collisions and aerodynamics.

DeepMind envisions uses ranging from training robots and agents to film previsualization and video game prototyping, as well as educational experiences, such as stepping into a historical painting and talking with its characters. The company acknowledges, however, that the model remains experimental: details can drift over time, text inside scenes is not always legible, and audio accuracy is still under development.

For now, Genie 4 is available as a research preview to a group of trusted testers, and it will reach Google AI Ultra subscribers in Project Genie and early-access partners of the Gemini API and Google AI Studio in the coming weeks. All video and audio it generates carries the invisible SynthID watermark, which makes it possible to identify AI-created content.

Photo: Google DeepMind

Source: Google DeepMind