Thursday, October 29, 2026

Google Announces Genie 4, an AI World Model With Sound. Only Testers Can Try It for Now

Google DeepMind's Genie 4 generates explorable worlds in real time, adds native audio and talking characters, and doubles world length to about two minutes. Here's what it is and what it means for you.

By Justin McBroom @JCredibleTech
· 2 min read
Image: Google DeepMind

Imagine typing “a crowded 1980s train station in rural Japan” and then actually walking down the platform, hearing the train, and having the station inspector talk to you. That’s the pitch behind Genie 4, the new world model Google DeepMind announced today. It’s also one of the most-talked-about tech stories on X and Hacker News right now.

Wait, what’s a “world model”?

A video generator like Veo makes a clip you watch. A world model makes a place you can move around in. Genie generates the world frame by frame, in real time, while you steer with normal keyboard controls. Think of it like a video game where nobody built the level ahead of time; the AI is making it up as you go.

What’s new in Genie 4

According to Google DeepMind and The Verge, the big upgrades are:

  • Sound, built in. Genie 4 is DeepMind’s first real-time world model that generates audio natively: wind, waves, music, and even characters who hold a conversation and react to what you do.
  • Change the world mid-scene. You can type things like a weather change or a new person or object while you’re exploring, and the scene adapts.
  • Longer worlds. Experiences can run about two minutes, up from about one minute for Genie 3.
  • A Gemini “director” in the loop. Gemini 3.8 Flash watches the scene as it’s generated, breaks goals into steps, and checks that each step actually happened, so the world stays consistent.
  • More realistic physics and people. Google says it handles things like collisions and aerodynamics better, and that people move and react more naturally.

What this means for you

Short answer: you can’t use it yet. Genie 4 is a research preview for trusted testers. Google says it will come to consumers and developers later, starting with Google AI Ultra subscribers in Project Genie and early-access partners on the Gemini API and Google AI Studio. For now there’s a waitlist.

Longer answer: Google is pitching this for robot and AI-agent training (letting them practice in simulated worlds before the real one), prototyping game ideas, and pre-visualizing film scenes. If that pans out, the first place most people feel it may be in the games and videos they watch, not in a Genie app of their own.

One honest caveat: everything we’ve seen so far are demo clips Google chose to show. We’ll know how good it really is once more people get their hands on it.

So would you rather play in a world an AI invents on the fly, or a game a human designed level by level?

Sources: Google DeepMind, The Verge. Images: Google DeepMind (Genie 4 demo stills).

TopicsAITechGoogleNews
All stories →