Worlds with sound on demand: Google DeepMind unveils world model Genie 4
Genie 4 generates interactive, photorealistic worlds in real time – for the first time with music, sounds and dialogue. Text prompts can change the weather or camera in the middle of a simulation.
Photo: Gciriani / Wikimedia Commons, CC BY-SA 4.0 (cropped and resized)
Video: YouTube · Google DeepMind · youtube.com
Google DeepMind on Tuesday unveiled Genie 4, the new generation of its so-called world model. The AI generates explorable, photorealistic scenes in real time and, according to DeepMind, is the lab's first real-time world model to also generate matching audio natively: music, ambient sound and dialogue. Users can change a running simulation by typing, for example switching the weather, moving the camera or adding people and objects. Scenes are meant to stay consistent for up to two minutes – twice as long as with its predecessor Genie 3. Google DeepMind's video shows examples.
Gemini models are part of the system: Gemini 3.8 Flash acts as a kind of director, planning multi-step events and checking whether each step succeeded. In a robotics experiment, DeepMind had a simulated Apptronik Apollo humanoid play checkers. For now Genie 4 is being tested only by selected partners – including filmmakers, photojournalists and game developers, for storyboards or game prototypes, for example.
DeepMind itself acknowledges the limits: over longer sessions or after many edits, details and objects can drift, and accurate audio and legible in-scene text are not yet assured. All output is watermarked with SynthID. In the coming weeks Genie 4 is due to reach Google AI Ultra subscribers in Project Genie and early-access partners of the Gemini API and Google AI Studio; Google did not give prices.