Google begins rolling out the Genie 4 world model in Project Genie to eligible AI Ultra subscribers
Google is bringing the Genie 4 world model to Project Genie, which runs in the browser, for eligible Google AI Ultra subscribers. Users can change a world mid-exploration by typing, but continuous use is limited to about a minute, according to heise online.
A world model does not produce a finished video. It generates the environment in real time as the user moves through it with the keyboard. In Genie 4, users can enter a new prompt mid-exploration to change the weather or add a character, for example, Google wrote in a blog post published on 18 November. The model also generates audio that adapts to events: ambient sound and character dialogue.
Genie 4 currently supports only about a minute of continuous interaction, the German site heise online reported. The new model card gives a resolution of 720p and a frame rate of up to 24 frames per second, heise said. Google says Genie 4 produces the highest-fidelity visuals, physics and audio of any of its models. For longer exploration, the company still offers Genie 3, in which a single world can now be explored for up to three minutes, up from 60 seconds. Genie 3 also lets users chain worlds together and start a world from a 360-degree panorama image.
Google said the models are still at the research stage. The company is still working on issues including camera latency, complex physical interactions, the accuracy of real-time audio and keeping a scene consistent over long periods.
In a behind-the-scenes video published by DeepMind, a representative of the Spanish studio SpecialGuestX says the same prompt works like a seed. A scene’s composition and action are the same every time for the first few seconds. That helps to reproduce a scene or keep a training environment under control, the representative said. This is a Google partner’s experience in Google’s own video, and it has not been independently tested. The video’s transcript is machine-generated.
The Project Genie agent uses Google’s models to create a world’s initial images and background music from a conversation. During exploration, Gemini can watch the screen, suggest directions and trigger events. Users can publish worlds to a community gallery, where others can explore and remix them. Google adds a SynthID watermark to all video and audio generated in Project Genie.
Access is opening in stages to eligible Google AI Ultra subscribers. The service is also available in Germany, and usage is subject to daily limits, heise online reported. Google AI Pro subscribers can join a waitlist on the Project Genie page.
Read more: Models