Imagine typing a few words and stepping into a living, explorable world, then changing the weather, moving the camera or adding a new character just by asking. That is the promise of Genie 4, the newest “world model” from Google DeepMind, which the company announced on Tuesday, October 27.

Here is what Google says Genie 4 can do, who can use it and why world models have become one of the most closely watched areas of artificial intelligence. The details below come from Google’s announcement.

What is a world model?

Most people know AI through chatbots that produce text or image generators that produce a single picture. A world model is different. It generates an interactive environment, frame by frame, that responds to what you do. Move forward and the scene changes as if you were walking through it. Take an action and the world reacts.

Think of it less like a video and more like a video game that is being invented on the fly, without a team of developers building every level by hand.

What’s new in Genie 4

Google describes Genie 4 as bringing “unprecedented realism and native audio to real-time interactive simulation.” The key features it highlights:

  • Natural-language steering. You can set an event in motion and steer scenes in real time by typing instructions, such as moving the camera, changing the weather or introducing new characters.
  • Native audio. Every world generates sound on the fly, including ambient sounds, music and interactive dialogue. Earlier world models were largely silent.
  • Built on Gemini. Genie 4 draws on Google’s Gemini family of models. Google says Gemini Omni anchors simulations in real-world appearance and behavior, while other Gemini models provide the reasoning needed to keep a simulation coherent and consistent over time.

That last point matters. One of the hardest problems for world models is consistency: remembering that a door you opened stays open, or that a building you walked past is still there when you turn around.

Who can try it

For now, access is limited. Google says it has started rolling out Genie 4 to a small group of trusted testers. It is “coming soon” to Google AI Ultra subscribers through Project Genie, and to early-access partners in the Gemini API and AI Studio. Google has also posted demos on the Google DeepMind blog.

Why Google thinks this matters

Google frames world simulation as “a crucial step toward AGI,” or artificial general intelligence. The reasoning goes like this: if AI can model how the world works, it can be used to:

  1. Train robots and AI agents safely. Virtual worlds let machines practice tasks thousands of times without breaking anything or anyone getting hurt.
  2. Power creative tools. Google specifically mentions filmmaking, journalism, gaming and manufacturing as industries that could benefit.
  3. Prototype ideas quickly. Designers could explore an environment or scenario before building anything physical or coding a full game.

The questions to keep in mind

As exciting as the demos may look, a few practical questions remain open until more people get hands-on time:

  • How long do worlds stay consistent? Real-time world models have historically struggled to maintain coherence over long sessions.
  • How will it be priced? Access through the AI Ultra subscription and the Gemini API suggests it will not be cheap to run at scale.
  • How will generated content be labeled? Realistic generated scenes and dialogue raise familiar concerns about misuse. Google has invested in its SynthID watermarking for AI-generated content, but it has not detailed how Genie 4 worlds will be marked.

The bottom line

Genie 4 is a glimpse of a future where you do not just watch AI-generated media, you walk around inside it and talk to it. It is still a limited research release, so most people will not be able to try it yet. But with native audio, language controls and Gemini under the hood, it is one of the clearest signs yet that interactive, AI-generated worlds are moving from lab demos toward real products.

Image: Let's Play! (Unsplash) by Glenn Carstens-Peters glenncarstenspeters via Wikimedia Commons, CC0 1.0