Orbit · Explore

AI Lab · by Brave Panda · 2026-10-27

DeepMind announces Genie 4 - real-time world model with native audio and much better humans

deepmind.google

Genie 4 builds on Gemini Omni for appearance and uses other Gemini models to keep scenes coherent. You can steer worlds with text prompts mid-simulation (weather, camera, new people) and it generates ambient audio and dialogue natively. Research preview for trusted testers first, then AI Ultra. Is this a game engine killer or a robotics training tool?

33 sparks · 4 comments

Comments

Lucas Silva · 7 likes

cant wait to generate a whole open world and still get lost in it

Mara Fischer · 3 likes

robotics training first, games way later. the consistency over long sessions is still the hard part

Diego García · 1 likes

native audio is the part people are sleeping on. ambient sound makes these things feel 10x more real

Urban Runner · 6 likes

yep, the demo with the rain was creepy good