Google DeepMind Says Genie 4 Builds Worlds That Talk Back and Change on Command
Google DeepMind's Genie 4 generates interactive worlds with native sound, talking characters and live text steering, and DeepMind calls it a step toward AGI. For now only trusted testers can use it, with Google AI Ultra subscribers next in line.
Type a sentence and a world shows up. Walk through it with your keyboard. Then type that you want a storm, and the sky turns while you watch, the wind gets louder, and the people standing around react to it. Some of them talk back. That is the pitch for Genie 4, the world model Google DeepMind announced on Tuesday. It is the first DeepMind world model that makes its own sound.
It is a striking demo reel. It is also, for now, mostly a demo reel. Nobody outside a small group of trusted testers can use Genie 4 yet, and every demo so far comes from Google itself.
DeepMind is not pitching this as a toy. In its launch post, written by Raia Hadsell, Diego Rivas and Hang Qi, the lab puts the AGI framing right up front.
"Understanding and simulating the world is a crucial aspect of building AGI." Google DeepMind
That is DeepMind's framing. What it actually showed is more concrete. DeepMind says Genie 4 delivers better physics, including aerodynamics and collisions, and much more natural human characters, which it calls the model's most dramatic leap. Users steer with standard keyboard controls and can type prompts mid-session to change the weather, move the camera or drop in a new person or object. Each world gets ambient audio generated on the fly, and characters can hold what DeepMind calls interactive dialogue that adapts to what you do.
Scenes stay coherent for about two minutes, according to The Verge and TechCrunch, double the roughly one minute Genie 3 managed. DeepMind has not published resolution or frame rate figures for the new model.
The interesting engineering is the babysitter. Gemini 3.8 Flash sits in the loop as a director, watching the scene, breaking goals into steps and checking incoming frames to confirm each step worked. Gemini Omni anchors how things look and behave. In one robotics experiment, DeepMind paired Genie 4 with Gemini 4 Argon, which split a task into steps that Genie rendered live, showing Apptronik's Apollo humanoid playing checkers. The idea is to rehearse robot scenarios in a simulated world before trying them in a real one.
Who actually gets to use it
Right now, trusted testers only. DeepMind says it has been working with cohorts in film, journalism, education, gaming and engineering. Filmmakers are using it for pre-visualization and game developers for quick prototypes. Rivas said the real-time part matters most in early creative work.
"The real time aspect becomes most valuable in the areas where keeping a state of flow in the creativity space becomes incredibly useful" Diego Rivas, Google DeepMind
Next in line, in the coming weeks according to TechCrunch, are Google AI Ultra subscribers through Project Genie and early access partners on the Gemini API and Google AI Studio. There is a waitlist. Ars Technica says developer pricing has not been announced and that the consumer rollout will come in phases on the Ultra plan, with cheaper tiers possibly later. If you are not paying for Google's top plan and are not a hand-picked partner, you are watching videos.
The Verge's Jay Peters was blunt about what that means for coverage.
"I haven’t been able to try Genie 4 for myself" Jay Peters, The Verge
DeepMind is unusually candid about the gaps, which is worth noting. It calls Genie 4 an early, experimental model and says visual details and object permanence can still shift or drift, especially in longer runs, after several complex prompt edits, or while moving around the scene. It lists complex physical interactions, real-time audio accuracy, legible text, spatial accuracy and long-term consistency as unsolved problems. Every clip carries a SynthID watermark across video and audio.
The competition is not waiting. TechCrunch notes Runway shipped its own audio-capable world model, GWM Worlds 2, a couple of months ago, while the sector consolidates fast. Two coherent minutes with talking people is a real step up from Genie 3. Whether it holds up once paying users start pushing it is the part Google has not shown yet.
