Google unleashes Genie AI that builds game worlds in seconds — here’s what you can do with it

Google unleashes Genie 3: Google DeepMind recently unveiled Genie 3, a game-changing advancement in AI-powered world models. This new version can create richly detailed, interactive 3D environments from a single text prompt in real time. This time, it will remember where you left off.

Google DeepMind's Genie 3 can support several minutes of consistent interaction and dynamic environmental changes, representing a significant breakthrough in simulations, agent training, and generative media.



ALSO READ: 7 insane new features coming to iPhone 17 Pro, and it’s launching next month

How does Genie 3 work in real time?

Genie 3 is Google's DeepMind's most advanced world model to date. It can create whole interactive environments from just a short text prompt. This version is different from the others because it lets you interact continuously for several minutes, and it runs smoothly at 720p and 24 frames per second.

People can walk through these AI-made worlds like they would in a video game, but none of it is pre-rendered. Instead, every frame is made on the fly based on what the user does.

Genie 3 outpaces its predecessor Genie 2 in nearly every key metric. Now, rather than just generating static scenes or short interactive clips, it enables sustained, coherent environments for several minutes at a time, as per a report by Google DeepMind.

ALSO READ: iPhone 17 release date confirmed, Apple promises its most powerful iPhone ever - here's what fans can expect

One of the most striking features is Genie 3’s visual memory. If you explore a generated world and then return to a spot you looked away from, objects and markings remain in place. This emergent capability, not explicitly programmed, lets Genie 3 preserve the continuity of your virtual space over a long horizon, sometimes up to a minute or more.

ALSO READ: Poop showdown! Furious farmers drench squatters with manure to reclaim their land - video goes viral

Genie 3 also adds “promptable world events”. Want to suddenly change the weather? Or spawn animals in a forest? You can update the scene on the fly with a new text command, and the virtual world adapts in real time

What can Genie 3 do?

Genie 3 can do a lot of different things. It can make natural settings, like rivers that flow, sunlight that changes, animals that act in different ways, and weather that changes. But that's not all. The model also uses fantasy and fiction to create imaginative settings and animated characters that can express themselves on demand.

There is also the option of historical and geographical simulation. Do you want to visit a Victorian city or an old market? You don't need any pre-made 3D assets to use Genie 3 to make things look like they did in different times and places.

You can even ask the model to change the weather, add new characters, or make things happen like a sudden storm, all from one line of text. These environments give researchers a lot of room to experiment when they are training AI agents, as per a report by Deepmind.

DeepMind has already started testing Genie 3 by putting its SIMA generalist agent into different worlds and giving it tasks to do, like finding its way around or collecting things, based only on what it hears from other people in the world.

DeepMind is very concerned about safety, ethics, and how these kinds of open-ended AI tools could be used wrongly. The preview is being run by their Responsible Development & Innovation Team, which wants to get feedback before making it available to more people, as per a report by Deepmind.

What are Genie 3's current limitations?

Genie 3 is still in the early stages, even though it has made progress:

• Time limits apply to interaction. Sessions that go on for a long time only last a few minutes before the accuracy of the visuals goes down, and the action space agents can control stays small.
• The dynamics of multiple agents are weak. The model has a hard time simulating complicated interactions between more than one moving character or agent in the same world right now.
• Accuracy in the real world is limited. Genie 3 can make settings that are similar to real or made-up ones, but it can't perfectly recreate real places.
• Text rendering is still very basic. If the input description doesn't say that clear signage or a readable UI is needed, it won't be.

How is DeepMind dealing with rollout and responsibility?

DeepMind knows that there are risks because it can create things in real time, like when immersive virtual environments are used in the wrong way or when algorithms have unintended biases. Genie 3 is being released as a limited research preview that only a small group of trusted academics and creatives can see. Researchers can get feedback and assess safety risks at this stage before deploying the system more widely.

From the very beginning, DeepMind's Responsible Innovation team is heavily involved to make sure that ethical safeguards are in place as they look into uses in education, entertainment, robotics, and agent-based systems.

The company is looking into ways to make access easier than just the current research preview. But Genie 3 starts a new chapter. In this one, AI doesn't just make text or pictures; it makes whole, living, interactive worlds that change and grow as you move through them.

FAQs

What distinguishes Genie 3 from earlier models?
Genie 3 supports real-time, multi-minute interaction in AI-generated worlds through visual memory, making it more consistent and dynamic than Genie 2.

Can anybody use Genie 3 right now?
Not yet; it is only available as a limited research preview to select academics and creators.