Genie 4 Opens to the Public in Project Genie
Google DeepMind
Three weeks after its research preview, Genie 4 reached the public in a rebuilt Project Genie, starting with Google AI Ultra subscribers in their browsers, with a waitlist for AI Pro. You can type a prompt mid-session to change the weather, drop in a character or redirect the action, with native audio that follows the scene, from ambience to dialogue. A Project Genie agent builds a world from a conversation, designing its look and scoring it, and Gemini can ride along, watching the screen and triggering events as you go. A community gallery lets people publish worlds for others to explore and remix. Genie 3 stayed in the model picker with longer runs: up to three minutes per generation instead of sixty seconds, worlds chained into longer journeys, and worlds seeded from 360-degree panoramas.
For filmmakers, this is the point where world-model previz stopped being a demo. DeepMind's own behind-the-scenes film showed the working method: build an environment from text, walk it with the keyboard, direct the action with live prompts, map character movement, dialogue and events to hotkeys with the agent, and reuse prompts as repeatable seeds so a scene can be run again for another take. Panorama grounding lets a real location become the start of a world, which is close to virtual scouting. Two of the launch worlds came from filmmakers and creators: "The Life of Ants," built with filmmaker Anand Gandhi, which explores a colony at ant's-eye scale, and "Villa Giallo," a room-by-room gothic mystery made with BLVCKL!GHT. DeepMind still listed the limits that matter on set: camera latency, complex physical interactions, how precisely the audio lands, and whether a scene stays consistent over time.