Chapter 10 — Animatic
Status: DRAFT 1 (2026-09-21), DESIGN v6 §4, §3 W4. Sources: docs/artifacts/STORYBOARD.md; docs/artifacts/ANIMATIC-VERDICT-2026-09-06.md; docs/artifacts/PREFLIGHT.md:60–64.
10.0 Objective
Given the recorded lines and the frame-zero stills, you can conform the whole film with the Chapter 1 tool, screen it cold, and revise the script and shotlist from what the viewer retold.
10.1 What you are deciding
Whether the film works, before a single video credit is spent on a card of the film (the stress takes of Chapter 9 were spent on the cast, not on the film). You have everything needed to find out: every line as audio (Chapter 7, with intention, Chapter 8), a still per card (Chapter 9), running times per card (Chapter 5), and a conform tool that joins stills at planned lengths (Chapter 1). The animatic is the whole film built from those, watched once by someone who has not read the script. Cards that sag get cut here, "not at D3" (PREFLIGHT.md:63), where cutting one costs the video that was generated for it.
10.2 How we did it
Frames. Thirty-eight storyboard frames, one per card, generated off-platform in an image chat from each prompt's first-frame, camera and lighting blocks (STORYBOARD.md:1–4), on 2026-09-05. "Nothing here enters the festival project. The frames exist to cut the animatic so the Director can watch the film once before a credit is spent" (:10–11).
Cut. 316 s, 38 clips at card runtimes, card id top-left, the line or beat burned in at the bottom, silent (:30–31). Scratch voice was planned (PREFLIGHT.md:61) and dropped in favour of burned-in line cards (:62).
Verdict (ANIMATIC-VERDICT-2026-09-06.md). "The Director watched the 316 s animatic and ruled card by card" (:3). Counted from the table: 26 keep, 9 shorten, 1 merge (3M into 3A, "the scene opens on 3A"), 1 cut (4A, the ring insert, "the ring is established at dinner"), and two restagings: 6E, "no kiss" (he lifts her chin, holds her gaze, 我必须走了), and 7B, "the sigh, not the ring" (a close on his face after the tail lights are gone, no hand insert). Result: 36 cards, 280 s, inside the 270–330 band (:49). The English exit line left the film; his last words became Mandarin (:70–72).
One consequence of the verdict shows what an animatic catches that a shotlist cannot: 7A had been 22 s of watching tail lights; the verdict shortened it to 16 s, which still breached the 12 s card ceiling, so it was split on a match cut into 9 s and 7 s with "no line in either" (:52–56). A hold that felt right on paper sagged on the screen.
Cost. Zero credits. Two days.
10.3 Case 10-A — what our animatic caught, and what it could not
- What I saw: the verdict fixed rhythm (nine cards shortened, two removed) and two stagings. Scene 5 still went on to produce three rooms (Chapter 5), 4C's line still landed half a second off its mouth (Chapter 11), and generated seconds were trimmed by about 71 % in the cut (Appendix A).
- Why I think so: the animatic was silent, so no line's real duration was tested against its card, and its frames were off-platform, so no returned room or face was tested. It tested the story's rhythm with the only two inputs it had.
- What I changed: the v6 animatic uses the recorded lines (durations are real) and the in-project frame-zero stills (the rooms and faces are the ones the video will start from). Both existed on HALFWAY by 2026-09-13; the animatic had been cut on 2026-09-05 and was never rebuilt.
- Would this happen on your tool? The gap is in the inputs, not the tool. Rebuild the animatic whenever the lines or the stills change; it costs a conform.
10.4 Do it with your tools
templates/build_cut.py takes a still as a source (held for the card's running time) and, per shot, an optional audio file with an audio_offset in seconds: the line file placed at the card's silence offset. One file per shot; a line that runs across a cut is a Chapter 14 matter. Tested 2026-09-21: a three-row table (still + line at 1.0 s, a clip, a silent still) conformed to 168 of 168 frames. If your conform tool cannot lay audio, play the line files from a second device while the picture runs; the point is the viewer, not the file. Burn the card id in only if the second reader is you; a cold viewer sees no ids.
10.5 Your turn
Time cap: one weekend. Spend: none.
- Build the shot table: every card, its frame-zero still (or the sheet crop if a card has no still), running time from the shotlist, the line file and its offset.
- Conform with the Chapter 1 tool. Decode. Ledger the frame count and duration.
- Watch it yourself once, impression first: one sentence of what you felt before any note.
- ★ Cold screening for the second reader with the viewing form (C.14): retell the turn, the slowest minute, who is who, one thing not understood.
- Rule card by card: keep / shorten / merge / cut / restage, with the new running time. Revise the shotlist and, where a line changed, the script and the recording.
- Rebuild the animatic from the verdict (a new cut folder; the verdict's running times in the shot table) whenever any card changed. Ledger both frame counts. Watch the rebuild with the second reader and ask only the slowest-minute question; they have read nothing more, but they are no longer cold, so the retell question is not asked again.
10.6 Gate (W4 — Animatic)
Each line needs valid evidence per DESIGN §5; outcomes are pass / fail / unverified, and unverified blocks.
- The animatic exists, conformed from the recorded lines and in-project stills, with its verification file (evidence: the files).
- ★ The second reader watched cold and the form is filled and initialled (evidence: the form).
- A verdict row per card exists; the shotlist's running times match the verdicts (evidence: the two files).
- The arithmetic still holds against the runtime floor after the verdict (evidence: the sum).
- No video credit has been spent on a film card; stress and test takes (Chapter 9) are not film cards (evidence: the ledger, job rows tagged stress/test versus card ids).
- Performance task: the rebuilt animatic exists with its verification file, and the second reader's slowest minute on the rebuild is either a different minute or the same minute with fewer seconds in it, measured from the two manifests: the cards inside the first form's slowest minute, their planned durations in manifest v1 summed, against the same cards' durations in manifest v2 (evidence: the two forms, the two manifests, your subtraction written on the second form; if the minute is the same and the seconds did not fall, the verdict did not act on what the viewer felt and the gate is failed).
Pass example from HALFWAY: the 2026-09-06 verdict, 38 cards ruled on in a table, two restagings that changed the ending, zero credits. Fail example: the same animatic never rebuilt once the lines and stills existed.