Genie 4 adds sound, keeps limits
Interactive worlds remain a research preview.
- Event / announcement date
- 2026-10-27
Google DeepMind’s October 27 Genie 4 preview adds sound to text-steered worlds. Trusted testers have access; Ultra subscribers and early API/AI Studio partners must wait. Broader access has no specified date or standalone price. [1]
Its predecessor, Genie 3, generated navigable worlds frame by frame rather than from an explicit 3D representation. DeepMind warned that errors accumulate during generation; its August 2025 report described consistency lasting minutes and imperfect geography. Those are historical specifications, not Genie 4 benchmarks. [2]
Creators can prototype scenes by adjusting cameras, lighting and dialogue in real time, DeepMind says. But objects still drift, and physical interactions and spatial accuracy need further research: convincing imagery does not establish dependable engineering simulation. [1]
The source
- Google DeepMind: Genie 4 preview ↗
Source date 2026-10-27T16:13:00+00:00 · reviewed October 29, 2026 · publication - Google DeepMind: Genie 3 research context ↗
Source date 2025-08-05T00:00:00+00:00 · reviewed October 29, 2026 · publication
About this item: News report. No first-hand reporting or independent testing is claimed. Corrections policy.
