By the Smili Phone team
Google DeepMind unveiled Genie 4 on October 27, 2026, and it is the most lifelike version yet of its world model technology. A world model is an AI that does not just draw a picture or play a finished video. It simulates a place, frame by frame, and reacts as you move through it. With Genie 4, you type a sentence or upload an image, and you get a photorealistic scene you can walk around in. Then you type another sentence, and the world changes while you are still standing in it.
Sound is the big leap this time
The headline change is audio. Genie 4 is DeepMind's first real-time world model to generate sound natively, meaning the sound is created in the same process as the picture, not dubbed on afterwards. A beach scene arrives with crashing waves. Music can play. Characters in the scene can talk back and react in real time to what you say and to what happens around them.
DeepMind says the sound is tied to events in the scene itself, so a collision sounds like a collision and a large hall echoes like a large hall. When the audio matches the action, your brain stops noticing the trick and starts treating the place as real, at least for a little while.
You steer the scene while it runs
The second change is live steering. While a simulation runs, you can type a plain language change: make it rain, move the camera, add a person, or drop in a new object. Gemini adapts your request to fit the scene on screen, and a guiding agent built on Gemini 3.8 Flash keeps planning each step and checking new frames so the world stays coherent. DeepMind also credits Gemini Omni with anchoring how people and places look and behave, and says Genie 4 handles physics such as aerodynamics and collisions better than before.
The third change is duration. Scenes now hold together for up to about two minutes, double the roughly one minute limit of Genie 3. Two minutes is still short, but it is long enough for a scene to develop: a character can cross a room, the weather can turn, a conversation can go somewhere. DeepMind is candid that this is still a limit, and details can drift over time, especially in longer runs or after several complex edits.
Who can try it, and what it is for
For now, Genie 4 is a research preview for trusted testers. Google AI Ultra subscribers will get it through Project Genie, and early access partners will get it through the Gemini API and Google AI Studio, with a waitlist open for everyone else. Google also says SynthID, its imperceptible watermark, will tag every video and audio clip Genie 4 produces.
It is worth keeping expectations in check. This is not a finished product you can download today, and it is not a replacement for a game engine. But it points to where interactive media is heading. If a world can be described in words, explored on foot, and changed mid sentence, the line between watching a video, playing a game, and chatting with an assistant starts to blur. Training simulations, classroom demos, and virtual tours are obvious early uses.
For everyday phone and computer users, the practical takeaway is simpler. The same AI progress that makes Genie 4 possible is already landing in the apps on your device, from smarter voice assistants to video tools that understand a scene instead of just a clip. If your phone is struggling to keep up with the newest AI features, bring it by our shop in Escondido. We can check whether the slowdown is a tired battery, full storage, or simply an older device.
Related: OpenAI models come to Google Cloud · Microsoft's Maia 300 AI chip
