docs/demo_script.md

AI Time Machine Demo Script

Use this path for judging or recording. It is designed to show the immersive layers in a reliable order.

Setup

• Adapter profile: fixture for fully deterministic local demo, or modal for live model demo.
• Coordinate mode: surprise, past, or future.
• Keep browser audio enabled. If narration or ambience is blocked, click once anywhere in the cockpit.

One-Take Demo

1. Open the app.
2. Choose surprise.
3. Click Launch.
4. Let the portal sequence run:
   • destination year counter moves from the current year
   • portal charges
   • world image fades in
   • character portrait resolves
   • narration caption/audio introduces the scene
5. Click Live voice or use text transmission.
6. Ask: Where am I, and what are you doing right now?
7. Ask: What do you think I am?
8. Ask: Show me something from your world.
9. Click Bring back souvenir.
10. Let the visual artifact reveal.
11. Close with: I come from another time. What would you ask me first?

Judge Talking Points

• This is not a chatbot skin. The model creates a whole encounter: place, ordinary person, voice, visual world, narration, and artifact.
• The 32B limit is respected per model:
  • Qwen/Qwen3-8B for language
  • Nemotron 0.6B for ASR
  • Chatterbox about 0.5B for TTS
  • FLUX.1 Schnell 12B for images
• The app remains a Gradio application and deploys as a Hugging Face Space.
• The experience is resilient: if image or audio generation fails, fixture/fallback visuals and captions keep the demo working.