Google announces Genie 4 AI model for interactive world simulations

In short: Google says its new Genie 4 AI model can generate longer, more realistic interactive experiences, and it can add audio.

What happened

Google announced Genie 4, which it calls a “world model.” That means it can simulate an interactive scene on its own, a bit like an AI that can sketch a tiny playable world instead of a single picture.

Google says Genie 4’s main updates are improved realism, the ability to change the experience while it is running using normal text instructions, and built-in audio generation. For example, someone could type a prompt to adjust what is happening in the scene, rather than stopping and rebuilding it.

Google also says Genie 4 uses Gemini 3.8 Flash to help run these simulations. Gemini is Google’s family of AI models, and “Flash” is a faster version designed for quick responses. Genie 4 can generate experiences that last about two minutes, up from about one minute in the earlier Genie 3 model.

Google is positioning Genie 4 as useful for film planning, early game idea testing, and training scenarios for robots. The company also notes a limitation, as simulations go longer, details can drift and objects may not stay consistent over time (like props moving around between shots).

Genie 4 is launching first as a research preview for “trusted testers.” Google says it will be coming soon to Google AI Ultra subscribers, and it has opened a waitlist for people who want to try it in Project Genie.

Why it matters

Tools like Genie 4 could make it faster to try out ideas for games, movies, or robot training without building everything by hand. But Google’s warning about details changing is important, because it suggests the tool may still need careful checking when accuracy matters.

Source: The Verge AI