Google's new AI can build playable virtual worlds, with sound

Google has announced Genie 4, an AI model that generates interactive virtual worlds, and this version can create audio to go with them. It is not open to the public. Google is launching it first as a "research preview" for "trusted testers."

What changed

Genie 4 is what Google calls a world model, meaning an AI that builds a simulated world you can interact with. Google lists four updates.

• Worlds look more realistic.
• Users can change a running scene by typing a plain-language request.
• The model can generate audio.
• It runs on Gemini 3.8 Flash.

Experiences now last about two minutes, up from one minute with the earlier Genie 3 model. Google pitches the tool for pre-visualization, which means rough mock-ups of a scene before filming. It also names game prototyping and training scenarios for robots.

What is not known yet

Google warns that in longer runs, details and object permanence can shift over time. In plain terms, things in a scene may change or fail to stay the same. The Verge's reporter has not tried Genie 4, so there is no hands-on verdict. No price or public release date has been reported.