Genie: Google's AI now creates games for you

O Google DeepMind has just released its newest project in the world of artificial intelligence: Genie, Generative Interactive Environments (Generative Interactive Environments), which promises to create truly playable games based on image prompts. See how it works:

What is Genie?

The new AI of Google creates games with controllable actions, as well as an entire interactive environment from just a photo or drawing. According to the company, this image can be of something that exists in the real world — such as an object — or created by some other AI tool that transforms text into an image.

This sets a great precedent for the future of the gaming world, as it would allow the creation of games with advanced controls in much less time, in addition to the possibility of decentralizing the process even further, giving amateur game creators more possibilities and leveling the competition. .

In fact, hypothetically speaking, the tool could go much further than that, generating instant virtual worlds, with hyper-realistic simulations.

The company also states that it will be possible to create games from drawings, from the most sophisticated ones made by professionals using tools such as Adobe Illustrator even drawings created by children or handmade sketches.

“In recent years we have seen the emergence of generative AI, with models capable of generating new and creative content through language, images and even videos…Today, we introduce a new paradigm for generative AI, generative interactive environments: Genie”.

Google DeepMind

How Genie creates games

O Genie It can create an entire controllable world with just one image, and according to models presented by the company, material with good definition is not even necessary for the AI ​​to work.

Trained from a set of online videos rather than stock labels, this new gaming AI from Google focused on learning 2D platform games and robotics on its more than 200.000 hours of data and trained 11 billion world models. Another notable function of the new Google Genie is to simulate deformable objects, and the company believes it could become a trainer for generalist AI agents in the future.

“…in an unsupervised way, Genie learns several latent actions that control characters in a consistent manner.”

Tim Rocktäschel, developer of Google DeepMind stated in Twitter.

Tim Rocktaschel, also stated in his account on Twitter ( X) that the AI Genie can create different trajectories from the initial frame and the focus of this creation is 'action' – a point that is the tool's biggest differentiator, which we haven't yet seen even in AI tools like Sora, also announced this month.

Tim also stressed that the learning process of Genie AI was done in a completely unsupervised video manner, using a type of temporally video-aware channelizer to compress videos into intuitive data units, a latent action (latent actions) that encodes transitions from 2 frames to one of 8 latent action and MaskGIT dynamics model that predicts future frames.

According to YouTuber Wes Roth in the video above, the latent action used makes a process of understanding what types of actions were taken between 2 frames of video. Therefore, if a character appears in one Frame on the ground and in the other he is in the air, he understands that the action that occurred between them is why he jumped. The dynamics model MaskGIT is used to predict what occurred after the action, that is, if a video game character jumps from a cliff, he will probably fall, continue his trajectory, but fall to the ground level from where he jumped.

Examples

A DeepMind and his collaborators even published some examples of prompts using your new artificial intelligence:

Availability

O Google Genie It is a tool under development and is not yet available to the public. The company did not provide a forecast for its arrival on the market.

This appears to be the most innovative AI tool created by the Google, as it can be the key to taking an idea off paper and transforming it into something concise and interactive in just a few moments. The new tool, according to what was presented in this first product demo, is what we expected from a technology giant like Google. Now we just need to know if she will live to exceed expectations!

If you are looking for more information about the world of artificial intelligence, we recommend this article which explains in detail what it is, the main types, advances in the area and more.

See also other features

Sources: Google DeepMind e X.

reviewed by Glaucon Vital in 27 / 2 / 24.