Google has developed an AI that can generate virtual worlds for video games. Called Genie 2, this new foundational model can create a playable scenario lasting up to a minute. Users only need to provide an image as a prompt, and the model will generate a 3D environment that can be navigated using a keyboard and mouse.
Please follow us on Facebook and Twitter.
According to Google, Genie 2 can simulate virtual worlds for all types of video games, from first-person shooters to driving environments or third-person gameplay scenarios. Developers at DeepMind state that their “world” model also simulates the consequences of any action performed within the virtual world.
The first demos of Genie 2 are impressive, showcasing gameplay of various types. The model can generate trajectories from a starting frame and remember parts that are no longer in view. This means that if the camera is moved to focus on another area, the AI will remember the previous view and render it once the camera is shifted back.
https://twitter.com/jparkerholder/status/1864314826891079787
In some cases, Genie 2 also generates a user interface, though this seems more for display than practical use. Other notable features include the ability to animate different character types, interact with objects—such as opening a door or shooting a barrel to make it explode—and even model non-player characters (NPCs) and simulate interactions with them.
The AI also considers elements like physics and special effects. For example, it can simulate water, fire, or lighting. If a character is riding a boat on a river, the boat will react according to the surrounding conditions. Other effects include reflections, smoke, and fog.
https://twitter.com/jparkerholder/status/1864314831970422878
Google’s AI Could Revolutionize Video Game Development
While AI’s role in game development has been widely discussed, no company has shown a model capable of achieving what Genie 2 can do. To create virtual worlds, Google’s AI relies on images generated by Image 3, its text-to-image model.
In theory, users could describe the world they want to play in natural language, select the image created by Image 3, and within 10-20 seconds, Genie 2 would generate the playable version. Google claims the virtual worlds remain consistent for one minute, ensuring that elements like the character and objects in the environment stay the same.
Another feature of this AI is its ability to create prototypes from a game’s concept art. A company like Ubisoft could use its art department’s designs to generate a playable prototype of games like Assassin’s Creed or The Crew, allowing them to test the concept’s appeal. According to Google, the worlds created by the AI support a maximum resolution of 1280 x 720 pixels.
Currently, this model is in an experimental stage and is not yet capable of creating a complete game. Like other AI, Genie 2 can sometimes “hallucinate,” generating content that doesn’t make sense within the world. Some animations are also choppy, and the designs still reflect the typical style of AI-generated images: saturated colors and a Vaseline-like filter that serves as anti-aliasing.