Genie 4 preview: AI worlds with sound you steer by text

Short answer: On October 27, 2026, Google DeepMind announced Genie 4, an AI "world model" that generates an interactive, explorable scene from a text prompt, complete with sound and speaking characters, and lets you change that scene while it runs. It is a research preview for "trusted testers" for now, with wider access coming soon. Everything it shows is generated by AI, and DeepMind calls the model early and experimental.

What happened

DeepMind describes Genie 4 as its new world model: you set an event in motion, and the model simulates what happens next, including how the world responds, how it sounds, and how the people in it move, react and speak. Music, ambient noise and interactive dialogue are generated together with the world; DeepMind calls it its first real-time world model to generate audio natively.

You steer the scene with keyboard controls plus text prompts. According to DeepMind, you can change the weather, move the camera or add a new person or object, and Gemini adapts your prompt to fit the scene. TechCrunch reports that a scene can stay consistent for up to two minutes, double the limit of Genie 3.

DeepMind also showed an experiment where a Gemini model breaks a task into steps that Genie 4 renders in real time, such as a simulated humanoid robot playing checkers.

Why people are talking about it

The leap is in how much is generated at once: more photorealistic scenes, more natural-looking people, matching sound and speech that reacts to what you do. TechCrunch notes it comes a little over a year after Genie 3 and calls it the most photorealistic and interactive version yet.

What it means if you use AI

• It shows where interactive AI is heading. Instead of only reading replies from a chatbot, you may one day step into a scene and talk to AI-generated characters. Those characters are AI, not real people.
• Testers see it as a draft tool. DeepMind says creatives try it for pre-visualisation and game developers for quick prototypes.
• Honest limits. DeepMind says details and objects can drift over time, and that complex physics, audio accuracy, readable text and long-term consistency are still being researched.
• Labelled output. DeepMind says all Genie 4 video and audio will carry SynthID, its invisible watermark for AI-generated content.

Quick checklist

1. Treat the clips you see as AI-generated demos, not footage of real places or people.
2. Check availability before getting excited: today it is a trusted-tester preview, with Google AI Ultra and Gemini API early access described as coming soon.
3. Join the waitlist on DeepMind's page if you want to try it.
4. Keep private details out of the prompts you type into any AI tool.

FAQ

Can I use it today? Only trusted testers can. DeepMind says broader access is coming soon.

Are the people in it real? No. Every character, voice and sound in Genie 4 is generated by AI.

Facts come from Google DeepMind's announcement and TechCrunch's report, both October 27, 2026. Plans can change, so check the official page.

Related reading:

• How to use AI as a personal assistant
• Siri AI now recalls past chats and adds 5 languages
• Study: 96% of ChatGPT memories were saved unasked

Cover image: a frame from the Genie 4 announcement video by Google DeepMind, generated by AI (SynthID-marked). Source: Google DeepMind.