Google DeepMind has unveiled Genie 4, a world model that builds an interactive scene from a text prompt and lets you move around inside it in real time. For the first time in the Genie series, those worlds come with their own sound: music, background noise and characters who talk back, according to Google DeepMind.
You cannot use it yet. Genie 4 is a research preview for trusted testers, and Google says it is coming soon to Google AI Ultra subscribers through Project Genie, its app for building and exploring these worlds. Here is what it does and what it would cost Australians to try it.
What Genie 4 actually does
A world model is an AI that generates an environment frame by frame and reacts to what you do in it, a bit like a video game invented as you play. You move with standard keyboard controls and steer the scene with plain text: change the weather, swing the camera, or drop a person or an object into the world.
Google says Gemini adapts each prompt to fit the scene around it, so a request lands in a way that matches what is already on screen. Genie 4 also simulates physics such as collisions and aerodynamics, and its characters move naturally and react to what happens around them.
Sound and dialogue are the big change
Google calls Genie 4 its first real time world model to generate audio natively. Music, ambient noise and dialogue are produced together with the picture rather than added afterwards, and characters can respond to what you say or to events in the world.
Several Gemini models work behind the scenes. Gemini Omni keeps scenes looking and behaving like the real world, Gemini 3.8 Flash acts as a guide that checks progress through a task, and Gemini 4 Argon breaks a goal into steps. Google’s demo uses this to get a humanoid robot playing checkers.

Who Google built it for
Google lists filmmakers using it as a draft mode to plan shots, game developers testing styles and mechanics in seconds, and teams turning a single archival photo into an explorable scene. Testers are also using it to simulate flash floods and wildfires.
Training robots and AI agents is the longer term goal. A simulated world is a cheap, safe place to practise a task before a machine tries it for real, and Google says business partners are testing it to generate scenarios from scratch.
When Australians can try Genie 4
There is no date yet. Access will open first to Google AI Ultra subscribers in Project Genie and to early access partners in the Gemini API and Google AI Studio, with a waitlist for everyone else. The announcement does not mention any country limits.
Google already sells AI Ultra in Australia from $149.99 a month, and its Australian plans page lists Project Genie, currently running Genie 3, as an Ultra only benefit. That makes Ultra the most likely first route for Australians, although Google has not said when Genie 4 will switch on here.
Why Australian businesses should care
For most businesses Genie 4 is a preview of where things are heading rather than a tool to buy this week. The clearest early uses are in marketing, training and design: walking a client through a space before it is built, or rehearsing a safety procedure on a simulated work site.
Every video and audio clip Genie 4 generates carries Google’s SynthID watermark, which matters for any business publishing AI content. And at $149.99 a month, Ultra only makes sense if your team will use it every week. Our ChatGPT vs Gemini guide compares the everyday plans, and deciding what is worth paying for belongs in your AI strategy.
The limits Google admits
Google is open about the gaps. Complex physical interactions, accurate real time audio and legible text are still being worked on, and scenes can drift visually over longer sessions. Expect impressive short demos well before anything a business would rely on for production work.
