Use this script for a 60–90 minute pilot focused on temporal coherence, drift, anchoring, and why video generation is harder than generating a single image.
Core question. What changes when generation has to work across time?
Materials
- Temporal Telephone
- Video Test Report
- Optional: A/B/C Comparison Board for coherence tests
- AI Use + Consent Checklist
- Optional: curated video clips or still-frame sequences showing drift and failure modes
Run of show
| Time | Segment | Facilitator move | Participant action |
|---|---|---|---|
| 0–5 | Welcome & bridge | Connect Session 2 to Session 3: a good frame is not enough; video must stay coherent across time. | Choose a participation pathway. |
| 5–35 | Temporal Telephone round 1 | Open Temporal Telephone in previous-frame-only mode. Use “a person at work” as the vague prompt, then check motion defaults after playback: job, body, setting, tool, pace, gesture. | Draw/save frames or observe where defaults and small changes accumulate into drift. |
| 35–50 | Temporal Telephone round 2 | Switch to anchor mode and repeat with the same basic subject or motion. | Compare whether identity, layout, and motion stay more stable. |
| 50–70 | Failure analysis | Use curated clips, still frames, or optional generated video examples. Name failure modes precisely. | Mark identity drift, physics breaks, camera inconsistency, style drift, or temporal jumps. |
| 70–90 | Video Test Report | Ask participants to complete one A/B/C or failure-analysis section. | Share one coherence claim and one responsible-use boundary. |
Optional pilot evidence move: use the A/B/C Comparison Board to document a hypothesis, one changed variable, a baseline motion prompt, one continuity anchor, and a stronger anchor or stress test. Keep real-person likeness and consent concerns visible before exporting.
Facilitator prompts
- “Where did the sequence first begin to drift?”
- “For the vague prompt ‘a person at work,’ what job, body, setting, tool, pace, or gesture appeared by default?”
- “What stayed consistent when only the previous frame was visible?”
- “What did the anchor preserve?”
- “What does video require that a single image does not?”
- “Which failure type is easiest to spot frame by frame?”
- “What would you need to specify to keep identity, motion, and setting stable?”
Investigation prompt
Compare a baseline motion prompt with one revision that adds an anchor: a consistent subject, fixed camera, repeated setting, or explicit continuity constraint. What improves, and what still breaks?
Low-AI / No-AI pathway
Participants can complete the session using Temporal Telephone and curated clips only. No one needs to generate video to investigate temporal coherence.
Fallback plan
- If drawing with the tool is awkward, have one facilitator draw while participants direct changes verbally.
- If playback is too fast or hard to see, step through thumbnails manually.
- If no curated clips are available, use Temporal Telephone outputs as the failure gallery.
- If participants do not want to analyze AI video, focus on continuity, consent, and classroom-safe critique protocols.
Pilot QA notes
During the pilot, note:
- Whether participants understood previous-frame-only versus anchor mode.
- Whether the Save frame and Play controls were discoverable.
- Whether thumbnails were large enough for discussion.
- Which failure categories participants used naturally.
- Any ethical concerns around generating motion, simulating people, or using real likenesses.