01What Google Genie 4 Is and How a World Model Works
Google Genie 4 is a new AI world model from Google DeepMind that builds a whole world from one sentence. You can walk around inside it in real time with a keyboard, like a video game.
DeepMind is Google's AI research lab, the team behind the Gemini models. A world model is different from an AI video. Once a normal AI video is finished, you can only watch it.
A world model builds a place you can move through.
Every second, Genie 4 works out what you should see next based on where you are going and what you are doing. Turn left, and it shows what is on the left. Walk forward, and the street keeps going.
- Nothing in the world is built by hand.
- There is no 3D artist and no game engine.
- The AI draws every frame live as you move.
You might type a snowy mountain, a busy street or a 1980s train station in Japan. Then you step inside. Google calls Genie 4 the most realistic and most interactive world model it has made, and both Sundar Pichai, the CEO of Google, and the head of Google DeepMind posted about it on launch day.
02What's New in Google Genie 4: Sound and Talking People
The first big change is native audio. Google says Genie 4 is its first real-time world model that makes sound natively. That means the same model creates the sound at the same moment as the picture, instead of someone adding it later in an editing app.
Every world gets sound that fits where you are:
- A forest gets wind moving through the trees.
- A beach gets the sound of waves.
- A train station sounds like a train station.
Google says every sound in its demo videos was made by Genie 4 itself.
The second big change is people who talk. Characters inside these worlds can hold a conversation. They react to what you say and to what happens around them, and the dialogue changes on the fly.
So you can walk up to someone in a scene and they will actually respond to you.
This turns a world from something you look at into something you interact with. Sundar Pichai described it as native audio matched to the world, from background sounds to characters you can talk to. Together, sound and speech make the scenes feel far more alive than earlier versions.
03Live Steering and Real Physics in Genie 4
The third new piece is steering in real time, the part people share the most. While a world is running, you type an instruction in plain English and the scene changes without stopping.
Here are the kinds of instructions you can give:
- Make it rain.
- Move the camera over there.
- Add a person.
- Add an object.
Google says Gemini takes each instruction and adapts it to fit the scene you are already in. Between your instructions, Gemini keeps watching and nudging the scene so the world moves forward in a way that makes sense. You move with normal keyboard controls and steer the story with text.
The fourth piece is physics and movement. Physics just means things behave the way they do in real life. Google says Genie 4 can simulate air moving around objects, collisions and other physical events.
The demos show this in action. One shows a watermelon sitting on a stand. Another shows a maple seed spinning as it falls inside a 1980s Japanese train station.
A third is a first-person ride on the back of a dragon. People in these worlds also move more naturally than before and react to what is around them.
04How Gemini Directs Every Google Genie 4 World
The fifth piece explains how all of this holds together. Google describes Gemini as a director for each world. Genie is the part that draws the world and makes it sound right, frame by frame.
Gemini is the brain that plans what should happen.
Here is how the director loop works:
- Gemini watches the scene.
- It breaks your goal down into steps.
- It checks the frames as they come in to see if each step worked.
- If the world changes or you add a new instruction, it changes its plan.
Google names the model doing this job as Gemini 3.8 Flash. Another model, Gemini Omni, helps keep these worlds looking and behaving like the real world. So Genie 4 joins several Gemini models that are already out and makes them work as one system.
Google's own post says this is where it thinks the future is going. Language models do the planning and reasoning. World models handle what you see and hear.
Both work together in one loop. That split is why you can steer a scene with text and still have it stay believable as it runs.
05Google Genie 4 vs Genie 3 and Project Genie
To see what changed, it helps to know the history. Genie 3 came out in August 2025. In January 2026, Google turned it into a product called Project Genie for people on its top plan, Google AI Ultra. Project Genie started in the US.
Genie 4 improves on that in several ways:
- Longer scenes: Genie 4 keeps a scene consistent for up to two minutes, which the episode describes as double what came before.
- Sound: the world now makes its own audio.
- Talking characters: people in the scene can hold a conversation.
- Live steering: you can change the world while it runs.
- Better visuals: scenes look much more realistic.
Keeping a scene consistent matters. It means the world you walk through stays the same world, instead of shifting as you move. That makes longer, more useful scenes possible.
Early testers are already posting about it, and some people are building games with this kind of tool. The jump from Project Genie to Genie 4 is about turning a short demo world into a place you can spend real time in and shape as you go.
06How to Get Access to Google Genie 4
Right now, Google Genie 4 is a research preview. That means Google is still testing it and only a small group has access.
Here is the rollout order described in this episode:
- A small research preview group gets it first.
- Google AI Ultra subscribers come next, inside Project Genie. Google says this is coming soon, and TechCrunch reports the rollout is planned for the coming weeks.
- After that, early access partners get it through the Gemini API and Google AI Studio.
If you want to be first in line, there is a waitlist for Project Genie linked from Google's announcement. Joining that is the simplest step you can take today.
While you wait, it is worth getting familiar with the other Google AI tools that sit around Genie, such as Gemini, Veo and Nano Banana. Julian Goldie's AI Profit Boardroom includes a 30-day roadmap for Google's AI tools, weekly coaching calls and daily tutorials, so members can see how to use Genie 4 the same week it lands for Ultra subscribers, without hours of testing alone.
07Real Uses for Genie 4 in Business and Beyond
Genie 4 is not just a fun demo. The episode covers several practical uses people and companies are already exploring.
- Film pre-visualization: a rough version of a scene filmmakers make before the real shoot. With Genie 4, you can run the same scene again and again and change the camera angle, lighting, dialogue and action live.
- Storyboarding: the comic-strip plan of a video, shot by shot. You can walk through a scene and try camera angles before you film or create anything.
- History: Google says photojournalists are turning a single old photo into a scene they can explore, and you could step inside a historical scene and chat to the people there.
- Game and product design: game developers try out looks, materials and how a game might work just by typing prompts, before writing any code.
- Company simulations: business partners are testing it to create new training and testing scenarios from scratch.
If any part of your business involves planning videos, ads or product shots, these uses are the most direct. You can test ideas inside a living scene before you spend money making the real thing.
08Why Google Genie 4 Matters for Robots and You
The biggest long-term use Google talks about is robots. In one demo, a Gemini model splits a job into steps and Genie 4 shows a robot doing it. One demo shows an Apollo humanoid robot.
Another shows a robot arm in a lab dispensing liquid into a plate.
The point is to test what a robot will do in a simulated world before it does it in the real one. That approach has three clear benefits:
- It is safer, because mistakes happen in a simulation.
- It is cheaper than testing with real equipment.
- You can run it as many times as you want.
Google also sees building worlds as a step toward AGI, the goal of AI that can do most things in the way a person can.
So who is Genie 4 for? Right now it suits filmmakers, video creators, marketers, game designers, trainers and anyone who plans scenes before making them. Once it reaches Google AI Ultra subscribers, business owners can use it to plan ads and product shots with no camera at all.
The more than 3,400 business owners inside the AI Profit Boardroom are already learning these Google tools so they are ready the moment access opens.