Google DeepMind has opened Genie 4, its most realistic and interactive world model yet, inside a revamped Project Genie experience. Announced on 18 November 2026, the update gives eligible Google AI Ultra subscribers first access; Google AI Pro members can join a waitlist from the same page.

Genie 4 turns a written description into an explorable 3D world — and for the first time, users can edit that world mid-session by typing a new prompt while they move through it: change the weather, swap a character, or restyle a scene on the fly. The model also generates native audio in real time, from ambient soundscapes to character dialogue, instead of layering sound on afterwards.

The relaunch adds a Project Genie agent that builds worlds conversationally, with custom soundtracks and live Gemini guidance, plus a community gallery to discover and remix other people's worlds. Genie 3 gets upgrades too: sessions of up to 3 minutes per generation, chained worlds, and 360-degree panorama grounding. All generated video and audio carry SynthID watermarking, and Google lists known limits around camera latency, complex physics and long-term scene consistency.

Genie 4 was first unveiled as a research preview for trusted testers in late October; this is its first step toward a wider audience. It is still early-stage technology — Google is clear it is a prototyping tool, not a game-engine replacement — but it is the clearest sign yet that interactive world models are moving from research demos into consumer products.

Sources