Google DeepMind Previews Genie 4, a Real-Time World Model With Native Audio and Mid-Simulation Steering
DeepMind's latest world model generates interactive, photoreal scenes with physics and sound — and lets users redirect the simulation with text while it runs.
Google DeepMind used earnings week to preview Genie 4, and the feature list suggests world models are crossing from research curiosity toward practical tooling. According to one roundup, Genie 4 is "a real-time interactive world model with native audio, text-based mid-simulation steering, improved photorealism, physics, and more natural human motion". A separate summary described it as a model that "generates real-time photoreal interactive scenes with physics, sound and dialogue".
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.