Overview
- Genie 4 adds native audio, more realistic physics and human motion, and lets users change scenes mid‑simulation by typing natural‑language prompts.
- The model uses Gemini 3.8 Flash as an active guide that breaks tasks into steps, evaluates generated frames, and steers simulations in real time.
- Genie 4 doubles short‑term scene consistency to about two minutes but still drifts after longer runs and has unresolved issues with object permanence, readable in‑world text, and perfect audio accuracy.
- Genie 4 was announced Tuesday, October 27, 2026, and is available now as a research preview for trusted testers with a waitlist for Project Genie access by Google AI Ultra subscribers and early partners.
- Google will embed SynthID watermarks in Genie 4’s generated video and audio, and the company positions the model for tasks such as film pre‑vis, game prototyping, and robot simulation while observers watch session length, compute cost, and broader rollout signals.