Google's Genie 4 Lets You Edit the World While You Walk Through It
Google DeepMind has opened Genie 4 to paying Google AI Ultra subscribers, turning a research preview into something you can actually poke at in a browser. Type a prompt mid-session and the weather changes, a character wa
Google DeepMind has opened Genie 4 to paying Google AI Ultra subscribers, turning a research preview into something you can actually poke at in a browser. Type a prompt mid-session and the weather changes, a character walks in, the action redirects — all without restarting. For the first time the model also generates native audio in real time, shifting between ambience and character dialogue as events unfold.
Here is the detail worth reading twice: the headline upgrade is Genie 4 going mainstream, but the concrete, quantified improvements belong to Genie 3, which stays in the model picker. Genie 3 sessions now run for up to three minutes, triple the previous 60 seconds, and it can chain worlds or seed one from a 360-degree panorama. Google published no session length for Genie 4 itself.
Key points
- Genie 4 opened to eligible AI Ultra subscribers on November 18, 2026
- Genie 3 sessions extended to three minutes, up from 60 seconds
- Real-time mid-session editing plus native adaptive audio
- Four admitted open problems: camera latency, complex physics, audio precision, scene permanence
- Original Genie trained on more than 200,000 hours of gameplay
- AI Ultra costs roughly 250 dollars per month, US only at launch
A world that forgets what it built the moment you look away is a demo, not a destination. Getting from that to a world that remembers is the whole ballgame for robotics and agent training.
Read the full story on threerouter: https://www.threerouter.com/blog/485
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.