Google has officially launched Genie 4, an AI model that generates real-time virtual worlds that users can move through themselves. [Photo: Google]

[DigitalToday reporter Jae-won Choi (최재원)] Google has made Genie 4, an AI model that generates virtual worlds users can explore themselves using only text or images, available to general subscribers. It also supports changing surroundings in real time and interacting with characters, advancing AI-based virtual world generation by a step.

On Nov. 19 (all times local), Japanese media outlet Gigazine reported that Google added Genie 4 a day earlier (Nov. 18) to Project Genie, its virtual world generation service. It is being rolled out sequentially to eligible users among Google AI Ultra subscribers.

A world model is AI technology that learns movements and changes in real environments and generates the next scene based on a user's actions. Users can describe a desired place or situation in sentences and then freely explore the virtual space like a game.

For example, if a user enters "a first-person view skiing in the Alps" or "a Japanese train station in the 1980s," it renders the scene. Users can move using the keyboard's WASD keys. They can also upload a reference image to specify how the virtual world should look.

A key point is that the environment can be changed in real time while exploring the generated world. If a user types, "Add a blizzard" or "Bring in wild animals," the AI reflects the changes in the scene.

Genie 4 also introduces real-time audio generation. Ambient sound, music and characters' dialogue are generated to match the unfolding scene. According to a model guide from Google DeepMind, it supports video output up to 720p resolution at 24 frames per second.

Google also introduced "Project Genie Agent," which builds a world while conversing with the user. If a user describes the desired mood, the AI composes an initial scene and background music. Completed worlds can be shared in a community gallery, or users can modify others' works to recreate them.

Genie 4 was first announced last month and was later provided as a research preview for a limited group of testers. The update expands access to general paid subscribers.

Still, there are limitations in precisely reproducing real physical phenomena and maintaining scene consistency over long durations. Google also cited improving complex object interactions and delays in camera response as tasks to address.

Google AI Pro subscribers can currently sign up for a waiting list, but no detailed expansion schedule has been disclosed. Google plans to develop world models not only for games and video production but also into technology for building virtual training environments for robots and AI agents.

Keyword

#Google #Genie 4 #Project Genie #Google DeepMind #Google AI Ultra
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.