Genie 4 is Google DeepMind’s newest world model, and it can now generate sound along with the visuals. It turns a text prompt into a playable 3D scene that you can steer in real time, but for now only trusted testers can try it.
Google released it as a research preview on October 27, 2026. Think of it as a very early look at where AI-made games and simulations are heading.
Here are the quick facts before the details.
- Genie 4 launched in research preview on October 27, 2026.
- It generates real-time audio, including music, ambient sound and dialogue.
- A scene stays consistent for up to two minutes, double Genie 3’s limit.
- Access starts with trusted testers, then Google AI Ultra subscribers in the coming weeks.
- Google hasn’t shared pricing or resolution details yet.
What is the Genie 4 world model?
A world model is an AI that predicts what happens next in a simulated space. You describe a place, and it builds one you can move around in, frame by frame, as you play.
Genie 4 is the latest in that line. TechCrunch reports it makes the most photorealistic and interactive worlds Google has released so far, with better physics and more human-like characters.
It isn’t a finished game engine. Think of it as a fast sketchpad for worlds, which is why Google also pitches it for learning and robotics.
What’s new in Genie 4
The headline change is sound. Genie 4 generates audio in real time that fits the scene, from background music to ambient noise to characters talking back.
The second change is steering. You can type a new prompt mid-simulation to change the scene, switch the weather or camera angle, or drop in new people and objects.
Scenes also last longer. A consistent world can run for up to two minutes, which is double what Genie 3 managed, per TechCrunch.
Here is how the main upgrades compare with their catch.
| Feature | What Google says | The catch |
|---|---|---|
| Real-time audio | Music, ambient sound and dialogue match the scene | Audio accuracy isn’t solved yet |
| Live steering | Change weather, camera, people or objects mid-play | Details can drift as the run gets longer |
| Longer scenes | Up to two minutes of consistent simulation | Still short for a full game session |
| Gemini as director | Gemini plans steps and checks frames while Genie draws | Research setup, not a product yet |
The catch column matters, because a demo that looks smooth for two minutes can still fall apart at minute three. That is the number to watch as Google keeps improving it.
Google says text inside the world is still hard to render legibly, and objects can wander off-model over time.
Who can use Genie 4 right now?
Only trusted testers, for the moment. Google says it will reach Google AI Ultra subscribers through Project Genie in the coming weeks.
Early access partners of the Gemini API and Google AI Studio are next on the list. That means developers may get hands-on before most consumers do.
No price has been announced, and the article doesn’t give resolution or frame rate. We’d treat any launch-date talk beyond “coming weeks” as a guess until Google says more.
Why Genie 4 matters beyond games
Games get the attention, but Google’s own examples go further. In one research experiment tied to Gemini Robotics, a Gemini 4 Argon model split tasks into steps, and Genie 4 simulated each one in real time.
That pairing is the interesting part. If an AI can rehearse a task in a simulated world first, it can fail safely thousands of times before a robot ever moves. You can read more about that model in our Gemini 4 Argon access guide.
If you like AI-made games, our guide to Google Playground shows the simpler, browser-based side of the same idea.
Trusted testers also see uses in film, mainly for ideas and storyboards. Google research director Tejas Kulkarni said we’ll “increasingly see in the future” hybrid world models, according to TechCrunch.
The field is also getting crowded. The same report notes AMD’s acquisition of World Labs and Amazon’s reported talks to buy Decart.
What to do now
If you’re a developer or creator, there’s one practical step: watch for Genie 4 in Project Genie and the Gemini API over the next few weeks.
Everyone else can wait. There’s nothing to download, and anything claiming to be a “Genie 4 app” today is not from Google.
One more thing worth knowing: research previews change fast. Features, limits and access rules can shift before a wider launch, so check Google’s own announcement before you plan around any of this.
Frequently Asked Questions
What is Genie 4?
It’s Google DeepMind’s world model that creates interactive 3D scenes from text prompts. This version adds real-time audio and live steering.
Can I use Genie 4 today?
Only if you’re a trusted tester. Google plans to open it to AI Ultra subscribers and some Gemini API partners in the coming weeks.
How long can a Genie 4 scene last?
Up to two minutes of consistent simulation. Objects and details can drift the longer it runs.
Does Genie 4 replace game engines?
No. Two-minute scenes and drifting details make it a prototyping tool for now, not a way to ship a full game.
My take: Genie 4 is impressive as a preview, but the two-minute limit keeps it a demo. The real story is robots and training, so judge it by what Google shows there over the next few months, not by the flashy game clips.
Leave a Reply