Google DeepMind’s Genie 4 can keep a generated world consistent for up to two minutes, double Genie 3’s limit, TechCrunch reports. It went into research preview for trusted testers on Tuesday, a little over a year after Genie 3.
New this time is real-time audio, including music, ambient sound and interactive dialogue. Users can also type prompts mid-simulation to change the scene, such as the weather, the camera angle or added people and objects. Physics and characters are said to be improved.
Gemini acts as a “director” that plans steps and checks frames while Genie generates the visuals. In a Gemini Robotics experiment, tasks were broken into steps that Genie simulated live, a possible way to test robot scenarios before real deployment.
Google pitches gaming, learning and robotics. Research director Tejas Kulkarni predicts hybrid world models that pair Gemini-generated code and text with Genie simulating reality at different levels of abstraction. Rollout to Google AI Ultra subscribers through Project Genie, plus early-access partners of the Gemini API and Google AI Studio, is promised in the coming weeks.
The limits are plain. Details and objects drift over longer runs, and Google has not worked out how to make audio accurate or in-world text legible. Engineer Diego Rivas called it “still early days”.
Two minutes is the number to watch. A robot test that needs ten minutes of steady physics is not covered by it, and the report gives no results from the robotics experiment beyond describing it.
