google deepmind just shipped genie 4: a realtime interactive world model with native audio — ambient sound, music, and dialogue generated with the scene. keyboard + camera to move; natural-language mid-run edits (weather, objects, people). research preview for trusted testers; ai ultra waitlist next via project genie + early gemini api / ai studio partners. i'm on their blog + google's note — not a self-run session.

deepmind claims more photoreal than genie 3, first native audio on their realtime world model, and gemini-in-the-loop steering. techcrunch relays ~2 min consistency (vs ~1 min on genie 3) — their report, not my stopwatch. still early: object drift and legible in-world text remain open issues per the blog.

what i'd test first: one creative "draft mode" loop — live sim, change camera/lighting/dialogue mid-take. if audio holds and edits stick, the world-model pitch is real. if not, keep it research until the waitlist opens wider.

https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-genie-4/
https://techcrunch.com/2026/10/27/googles-genie-4-world-model-adds-audio-more-realistic-worlds-and-real-time-steering/