AI news · Google DeepMind · Genie 4
Google’s Genie 4 builds steerable worlds with sound
DeepMind’s new world model makes interactive scenes you can redirect by typing, holds them for two minutes and adds music and dialogue as it goes.
Google DeepMind announced Genie 4 on Tuesday, 27 October 2026. It is a world model, an AI that builds a scene you can move through while it runs, and you change it as it plays by typing a request for new weather, a new character or a new camera angle, Google’s post says. Sound comes with it: background noise, a score and characters who talk.
Scenes hold together for up to two minutes, against one minute for Genie 3, TechCrunch reports. Gemini now runs the show: it breaks a request into steps and reviews each frame Genie produces, while Gemini Omni sets how things look. Trusted testers get it first; Google AI Ultra subscribers come next in Project Genie, along with early partners on the Gemini API. Google’s pitch covers robot training, games and education.
Ultra is Google’s top plan. In our 27 October check, it costs $99.99 a month with 10,000 Flow credits, or $199.99 with 25,000, while a 720p Gemini Omni Flash clip in Flow costs 7 to 15 credits on any plan.
Genie 4 builds on Gemini Omni, the model that already makes 720p Flow clips for 7 to 15 credits on any plan, and Genie itself comes to subscribers through the $99.99 or $199.99 Ultra plan, so for most ad teams a short clip still comes from Flow.
Sources
Alexandre S.
Alexandre S. is a creative ad strategist and copywriter with ten years' experience in digital marketing across multiple channels.