Google DeepMind's Genie 4 builds 3D worlds you can walk through, with sound and talking characters

What is Genie 4?

Genie 4 is a "world model," an AI that generates a scene and keeps simulating it as you move around, rather than rendering a fixed video clip. Google DeepMind announced it on Tuesday, October 27, 2026. You steer with normal keyboard controls for walking and the camera, and you can type a prompt mid-scene to change the weather, move the camera or add a new person or object.

The big new trick is sound. Google DeepMind calls Genie 4 its first real-time world model to make audio natively, meaning the sound is generated along with the picture. Each world gets its own background noise and music, and characters can talk and react to what you do. According to Google, it also handles physics, such as collisions and things moving through the air, and human movement more realistically than before.

Behind the scenes, Google's Gemini models do the planning. A Gemini model (Gemini 3.8 Flash, per Google) watches the frames as they're made, breaks a goal into steps and checks that each step worked, so a scene stays on track.

How is Genie 4 different from Genie 3?

According to TechCrunch, Genie 3 arrived as a research preview a little over a year ago, and Genie 4 can keep a scene consistent for up to two minutes, double Genie 3's limit. It adds native audio and talking characters, looks far more photorealistic, and lets you make free-form edits while the world is running.

When can I use Genie 4, and what will it cost?

Not yet, for most people. Right now Genie 4 is a research preview for "trusted testers," the small group of partners who get early access to Google's experimental tools. Google says it's coming next to:

• Google AI Ultra subscribers, through Project Genie, Google's tool for making interactive worlds from text or image prompts.
• Early-access partners on the Gemini API and Google AI Studio, the routes developers use to build with Google's models.

TechCrunch reports that rollout to those groups should happen "in the coming weeks," while Google's own post just says "coming soon." Google hasn't announced a price or a date for wider access. There's a waitlist on the Google DeepMind blog. All video and audio from Genie 4 carries SynthID, Google's invisible watermark for spotting AI-made content.

What could Genie 4 mean for creators and small businesses?

Google says the full product is still unclear. Its trusted testers are already trying uses that sound a lot like everyday creative work:

• Storyboards and "draft mode" for video. According to Google DeepMind, filmmakers are using it to step into a live scene and run several "takes," changing camera angles, lighting, dialogue and action as they go. Google engineer Diego Rivas told reporters the real-time part matters most "in the ideation phase, in the storyboarding phase."
• Game and experience prototypes. Game developers are exploring how fast they can go from an idea to something explorable, trying out looks and mechanics with plain-English prompts before writing any game code.
• Training robots and AI agents. This is Google's bigger goal: safe, endlessly varied practice worlds for machines before they meet the real one.

For a solo creator or a small brand, the near-term picture is pitching and planning. Picture walking a client through a rough version of an ad's setting, with sound, before anyone books a location. That's a guess at where this goes, and Google hasn't promised any of it as a product feature.

What can't Genie 4 do yet?

Google lists the gaps itself. Details and objects can shift or "drift" the longer a simulation runs, after lots of prompt edits, or as you move around the scene. Complex physical interactions, accurate audio, readable text inside the world, spatial accuracy and long-term consistency are all still being worked on. So anything with a logo, a price tag or a street sign is a bad fit for now, since the letters may come out as alphabet soup.

What people said on X

Google DeepMind's launch post passed about 2.4 million views by Wednesday morning. CEO Demis Hassabis followed up that evening with the waitlist news.

What to do this week

1. Join the waitlist only if you're curious. It's linked from the Google DeepMind blog post. Joining costs nothing and doesn't commit you to a plan.
2. Don't upgrade to Google AI Ultra for this yet. Google hasn't said when Ultra subscribers get it or what it will cost to use.
3. Video and ad makers: jot down one storyboard or pitch you'd love to "walk through" before filming. That's the job to test first when access opens.
4. Everyone else: nothing to do. Check back when Google announces wider access.