Create Universes That You Can Navigate With A Few Words: Google Changes Everything Genie 3

The things you see was taken from a game. Was retrieved from a source image. The things you see in real-time universes generated by artificial intelligence.

This week in the world of artificial intelligence, developments, there were more frantic than ever. GPT-5 output X output grok imagine, version 4.1 output Claude's opus Eleven Labs made the experience of composing music. 3, which is among the most exciting developments was the Genie but all these DeepMind Google. Because this technology is quite different from the others.

Just like the others text, it doesn't produce visual or video. Entirely within navigable, is creating interactive worlds. Just a few sentences you write and the system for you at 24 frames per second at 720p resolution, produces dynamic environments that are able to maintain the consistency of several minutes.

 Genie 3 How Does It Work? Summer and travel

Pre-produced at the Instant is being modeled

The images that are currently described in these lines pre-prepared with a 3D modeling program. A game engine such as Unreal Engine was not used. While you move in by typing was created and continues to be created. The same games as in the world, back and forth, left and right can move. You can turn your head and look around. As you move, continues to create new places.

When interacting with the world — for instance, when I started painting the walls — that part is reproduced by your interaction. As the name implies, this system currently 3. has reached version. DeepMind Google for more than 10 years of continuing research on environments and simulation of the system that was born as a result.

The product of 10 years ' research

In the past there were more primitive models. At that time, called the agent, agents, artificial intelligence, real-time strategy games they were training in. Now to be able to create entirely new worlds had become.

The Dynamics Of The Physical World: The Oceans Volcanoes

Realistic physics simulation

Genie is one of the most impressive features of the complex dynamics of the physical world 3 how much can be realistically simulated. A volcanic terrain in challenging terrain conditions from the perspective of a wheeled robot, we see how it is progressing. While navigating on the wheels of the vehicle, away from the black rocks of the volcano, smoke, have details such as lava flows.

The dynamics of the water when using a jet of water at a festival, we can see the reflections of the light. We use the created object in front of vehicle's headlights fall on top of the can, and we can multiply to that object. Also the dynamics of the physical world is being created — it's not just the visuals. And all is done without any modeling or coding.

The level of detail and natural ecosystems

Simulated an approaching hurricane in Florida during a walk by the sea — wind, and the waves hit the road after the remaining traces of moisture you can feel. A small detail: that doesn't seem much affected from the wind in the palm trees — the system still has some shortcomings.

Tree on the shore of glacial lake in the sunny environment as they headed out of the shadow like when the exposure value of the camera is changing. Obviously, the properties of glass that is used in such environments artificial intelligence flaws and learned a lot even the action reflected their environment.

Okyanusun among hundreds of jellyfish can swim in the depths — biyoluminesan lights, deep-sea canyons, all of them incredibly realistic. In the example of a Japanese zen garden of the morning light fell on patterns drawn in the sand, pond, lily pads, moss-covered rocks, bamboo fence — everything in photo-realistic real-time.

The world of gaming and creativity: endless Level, Endless Character

Isn't the real world, and there's a world of dreams

Genie 3 only simulate the real world it doesn't. We can imagine worlds can create a completely different logic. Running over the Rainbow Bridge, cuddly creatures, created in the style of 3D animation cap universes with vibrant colors origami paper folded in a world of advancing in the style of arcade games and theoretically it is possible to produce an infinite level.

The only level in the same way to the limit with the eternal character can be produced. As a huge Firefly fly in a magical forest — tree lighted houses, cottages, a magical atmosphere, it was out of a fairy tale.

Size storytelling

Also what happens if it is added to the size of storytelling? A book can be nested lived experience of the game. Reading the Lord of the Rings, as well as navigate to Rivendell, and you can imagine being able to interact with the characters there.

Google Maps + Genie 3: a voyage through history and virtual Tourism

With the potential to merge maps

There are Maps and Google Earth applications — we are able to go down to the streets and look around, all the cities of the World, significant natural areas, with the scale and real cases processed. You imagine this combined with artificial intelligence. 3 Genie can't do it right now, but similar can uredebi. An environment that was created from the influence of the Alps — kayakil slopes, narrow passageways, Yesil vegetation. From Venice, in an environment that was created under the influence — the texture of the duct around the old buildings, boats, gondolas.

Revive the past

Knossos Palace in Crete — currently in the real world, however, we are able to wander among the ruins. But here we are able to discover him with the Golden Age. To be able to make a journey into history, offer huge advantages in the area of Education. Never go to the places where you should go or you will refrain able to convert the opportunity to experience.

Memory and consistency: the dream like to see

The perception of space and the token limit

Tokenlar is measured in artificial intelligence, knowledge, and capacity. When producing each frame that was previously created automatically, you have to keep in memory the whole scene. A minute later, back to the same place needs to remember the previous state when you get a minute — or you will start to lose your perception of space.

House painting we can see it in the example of: hand while painting the house after looking in other directions with a roller, paint it needs to be a part of. The model is currently on a short-term such information as he can remember. When a magical portal opens and entered the portal in a neighborhood of Victorian style homes, you're not being transported to a desert — both media also continues to be simulated in a consistent manner.

Memory test in a classroom environment

There is written on the blackboard: "Genie 3 memory Test". Apples Cup and drawings. Turn your head the other way and looked out of the window continues to remain writings and drawings on the board and back again. If you look for more than a minute on the blackboard, written in the model there-is starting to forget the things that are underlined. In a sense, like in a dream.

This is a feature the developers of spontaneous coherence — emerging capability — it is. Also you can create or Gaussian Nerf splatting consistent alternative methods such as 3D environments, but they prepared in advance for 3D modeling needs. 3 genie of the creation of the world much more dynamic — each frame on the basis of the text and the actions the user is being created from scratch.

"What If?": Change The Script Instant

Instantly convert text command with the world

The text has the ability to change the world instantly provide you with commands. Just don't use the navigation controls — can change the weather conditions of the environment, new objects, new characters you can add or remove.

This feature is very impressive and why? Because there's a pattern in storytelling: ask "What if?" — and this allows you to branch. Geciversey was on the street while visiting London or a jetski your side? What if you only had a runner or in a chicken suit? Or what if you had a dragon? These alternative scenarios imagination is not only valuable in terms of other models can be used to train the artificial intelligence.

Integration with Sima agents

Google DeepMind, Genie SIMA 3 agents tested. SIMA is a general purpose agent that can run in 3D virtual environments. Every agent in the world is given 3 different navigation commands the genie is sending those goals and to reach this goal. Thanks to the consistency of the model, much longer and complex targets, it has become possible to create task sequences.

Limitations and future: what time will be open to everyone?

Existing restrictions

Only a few minutes continuous interaction can be supported currently — you are in a world of hours and I don. It's like a dream walk: a few minutes ago and the place went up in the middle and is being rebuilt. Real-world locations are geographically accurate, cannot be simulated. And most importantly: the technology yet are not open to everyone. It's not enough for 720p resolution with virtual reality glasses — minimum 4K, 8K, and need the ideal in a stereoscopic image.

Access and future forecast

Google DeepMind announced as a preview of this technology is limited in terms of security and responsibility for the research. Just a little scholar and early access to the creative group is provided. They will gather feedback, they will know the risks, and then they say that they're going to open up access to everyone. They don't give an exact date full, but when we think about Google's other technologies — they started — not years, months later, that we could have access to this technology is able to predict.