# Kiosk — Process
**Role:** the premiere surface. Full-fidelity character (Cine assembly, strand hair), live Q&A, authored programs. See [[Specification — Full-Body Historical Virtual Human]] §2.1.
## Two architectures
| | Local PC | Pixel Streaming 2 |
|---|---|---|
| Rendering | RTX 5080–5090 class PC in the enclosure | Cloud or on-site GPU server streams to a thin client/browser |
| Pros | Lowest latency, works offline | One GPU can serve several screens; easier updates |
| Cons | Hardware per kiosk; heat and noise | Needs reliable network; per-hour GPU cost |
| Fit for Nantucket | **Default** (Atheneum, MMA) | Later, for many screens at one venue |
## Display options
Portrait 55–86 inch display · LED blade · holographic cabinet (e.g. **Proto Luma**: 86-inch, built-in camera, 6-mic array with echo cancellation; can show a connected PC's output). Confirm whether a cabinet vendor's platform allows your own Unreal build and your own AI stack before buying.
## Steps
1. Site survey: power, ventilation, sightlines, ambient noise, network, ADA reach ranges.
2. Choose display and enclosure; mic array + push-to-talk; presence sensor for approach mode.
3. Build: Windows kiosk image, auto-login, app auto-start, watchdog, remote management, nightly restart.
4. Modes: **rest** (idle life) → **approach** (90-second greeting) → **program** (monologue) → **Q&A** → **out of character**.
5. Offline fallback: authored monologues and captions only.
6. Privacy: presence detection only; log questions, never faces.
7. Soft open with docent; daily checklist; monthly log review.
**Owners:** Unreal engineer, kiosk/AV integrator, venue ops. Tutorials: [[Kiosk — Tutorials]]