Google's Genie 4 builds worlds with their own sound, but almost no one can get in
Google DeepMind Genie 4 simulates scenes in real time with native audio and text control. Today only a few can try it.
Genie 4 is Google DeepMind's new world model: it simulates scenes in real time, generates its own sound, and obeys text instructions. The catch is that, for now, only a limited group of trusted testers can try it.
What is Genie 4
Google DeepMind introduced Genie 4 on October 27, 2026. It's a world model: instead of producing a closed clip, it simulates a scene in real time. The underlying idea is that directing a scene no longer requires filming it.

What Genie 4 does
Three things stand out:
- It generates its own sound: music, ambience, and dialogue, natively.
- It obeys your text: you can change the weather, move the camera, and add characters with written instructions.
- It keeps the scene coherent, and it isn't limited to a short clip: according to TechCrunch, it lasts two minutes straight.
Instead of building a physical set, you describe the scene and the model builds it.

Who can use Genie 4 today?
Almost nobody. For now only a limited group of trusted testers is trying it. For everyone else, the promise is that it will reach Ultra soon, but with no confirmed date. Treat the access claim with caution: what exists today is an announcement and a restricted trial, not an open product.
How to apply it in your business?
If your team produces video, advertising, or content, Genie 4 points to a future where previewing a scene is as simple as writing it. Until there's general access, the useful move is to prepare:
- Save this post for when it reaches Ultra.
- Identify which scenes in your productions currently require building a set, and which you could describe in text.
- Share it with whoever makes video on your team.
Another recent case of restricted access from Google is Gemini 4 Argon: read the full breakdown of Gemini 4 Argon for defenders only.
Frequently asked questions
What is Genie 4?
A Google DeepMind world model that simulates scenes in real time, with native audio and text control.
When was Genie 4 introduced?
On October 27, 2026.
Does it generate sound?
Yes: music, ambience, and dialogue.
Can I use Genie 4 now?
Not generally: for now a limited group of testers is trying it. It's coming to Ultra soon, with no date.
Conclusion
Genie 4 shows how world models are moving from clips to scenes you direct with text and that sound on their own. Almost nobody can get in yet, so save it for when it reaches Ultra and send it to whoever makes video.