Google DeepMind Genie 4: the AI that creates playable worlds in real time, explained
Genie 4 creates interactive 3D worlds you can walk around in, with sound, realistic physics and people, and lets you change the scene with text as you play. Here's what it does and when you can try it.
Imagine typing “a rainy street market at night” and then walking around inside it, hearing the rain and the crowd, and changing the scene as you go. That’s what Google DeepMind’s Genie 4, announced on October 27, 2026, can do.
What is a “world model”?
A regular AI video tool creates a clip you watch. A world model creates an interactive world you can move through. It predicts what should happen next as you act, a bit like a video game that’s generated on the fly instead of being built by developers.
What Genie 4 can do
- Explore with a keyboard: move around generated worlds using normal keyboard controls.
- Change the world with text: type prompts mid-scene to change the weather, the camera angle, or add people and objects.
- Sound included: DeepMind says it’s the first real-time world model to generate audio natively, including ambient sound, music and spoken dialogue.
- More realistic physics and people: better collisions and movement, and characters that react.
- Longer consistency: scenes stay consistent for up to two minutes, double the previous version (Genie 3), according to TechCrunch.
Gemini as the “director”
Genie 4 works with Google’s Gemini models. Gemini helps plan what should happen, interprets your edits so they fit the scene, and checks that each step makes sense, while Genie generates the visuals and sound in real time.
Genie 4 is less like watching a video and more like stepping into one, and being able to rewrite it as you go.
What could it be used for?
- Games: quickly prototyping game ideas and levels.
- Film and design: planning scenes and storyboards before filming.
- Education: exploring historical places or turning old photos into explorable scenes.
- Robotics: testing robot tasks in simulation before trying them in the real world. DeepMind showed an experiment with a humanoid robot playing checkers.
Limitations
DeepMind is clear that Genie 4 isn’t perfect:
- Details can drift or objects can change during long sessions or after many edits.
- Complex physics, readable text and audio accuracy still need work.
- It’s an early research product, not a finished consumer app.
Can you try Genie 4?
Not yet for most people. It’s currently a research preview for trusted testers. Google says it will roll out in the coming weeks to Google AI Ultra subscribers through Project Genie, and to early-access developers through the Gemini API and Google AI Studio. There’s a waitlist.
Is it safe?
DeepMind says Genie 4 went through internal safety checks, has guardrails, and that all generated video and audio carry an invisible SynthID watermark to help identify AI-made content.
The competition
World models are a hot area. Runway released a playable world model with audio a few months ago, and other world-model startups have attracted interest from big tech companies.
Related: Google’s Gemini app hits 1 billion users.
Frequently asked questions
What is Genie 4?
A world model from Google DeepMind that generates interactive, explorable worlds in real time, with audio, from text and images.
Is Genie 4 a video game?
Not exactly. It generates game-like worlds on the fly, but it’s an AI research tool, not a finished game.
How can I use Genie 4?
It’s limited to trusted testers now. Google plans to offer it to AI Ultra subscribers and developers soon, with a waitlist open.
What’s new compared to Genie 3?
Native audio, more realistic physics and people, real-time text edits, and scenes that stay consistent for up to two minutes.
