Models
Google's virtual worlds now have sound and change on command, but almost no one can get in
Genie 4 creates more realistic scenes you can explore, with their own sound, that change with a single sentence. For now, only invited testers can try it, and Google admits the picture gets worse over time.

The essentials
- On October 27, Google announced Genie 4, which generates explorable virtual worlds with their own sound: ambient noise, music and dialogue (Google Blog).
- While the scene is running, you can change it by typing: move the camera, change the weather or add characters (Google Blog, TechCrunch).
- It keeps a scene consistent for 2 minutes, twice as long as its predecessor, Genie 3 (The Verge, TechCrunch).
- For now, only testers chosen by Google are using it. It will soon reach Google AI Ultra subscribers and some developers (Google Blog, TechCrunch).
Why it matters
World models have become a priority for AI labs (Ars Technica), and the industry is starting to consolidate: according to TechCrunch, AMD is buying World Labs and Amazon is in talks to buy Decart (TechCrunch). Genie 4 suggests Google wants to stay ahead. Its clearest short-term use seems to be trying out ideas: film sketches, game prototypes or robot training (The Verge, TechCrunch).
The details
A world model creates a virtual environment frame by frame as you move through it, without anyone having programmed it beforehand (Ars Technica). According to TechCrunch, Gemini acts as the director in Genie 4: it plans the steps and reviews the images Genie creates (TechCrunch). In a robotics experiment, Gemini 4 Argon broke tasks down into steps that Genie 4 simulated on the spot. It is a way to test a robot before taking it into the real world (TechCrunch).
Google acknowledges the model has limits. According to what its team told TechCrunch, details and objects drift the longer the simulation runs. They also haven't managed to make the sound accurate or the text inside the world readable (TechCrunch). Engineer Diego Rivas stressed that it is early days and that it is not yet clear what the final product will look like (TechCrunch).
What we don't know
- When exactly it will reach Google AI Ultra subscribers, and whether it will be available in Spain.
- No one outside Google has yet checked how much better it is than Genie 3.
Sources
- We’re announcing Genie 4, our new world model for real-time interactive simulation.
- Genie 4: A new generation of world simulation
- Google’s next Genie AI model generates more realistic experiences
- DeepMind unveils Genie 4 world model promising more realistic physics and humans
- Google’s Genie 4 world model adds audio, more realistic worlds, and real-time steering
- Meet Google DeepMind’s Genie 4: Real-Time World Model With Native Audio
- Google’s AI worlds can now talk back
- Import AI 479: Spacewar; low-latency tasks versus LLMs; Genie 4
Was this story useful? Yes Not really