Pipeline Hub

format-mini-movie

Long-form animated mini-movie ad, 5-12 min in CHAPTERS: the extended Dawn recipe A (accused mom, vindicated) with escalating chapters, one subplot, a mid-point false solution, the explainer, the reveal, a long payoff and the offer, spoken by a mom narrator plus voiced characters.

Lane: cartoon-h3 dialogue (one pack, one run) · not built yet: 9 (see below) · source: /Users/ayden/.openclaw/workspace/skills/format-mini-movie

SKILL.mdbeat-sheet.mdsources.mdEvals (6)Learnings (13)
datelearningsource
2026-10-06No cross-run chapter concat exists: assemble.py joins one run's batch clips under one VO, one bed, one caption track, one end card. Build a mini-movie as ONE pack / ONE run; regenerate a chapter with --batches/--regen inside it.scripts/cartoon_h3/assemble.py fit/concat.txt
2026-10-06Forced batch breaks exist only at lead_start/body_start/body_end; a chapter seam can fall inside a batch (becomes a named Transition). A chapter_break key is not built.scripts/cartoon_h3/storyboard.py build_beats/build_batches
2026-10-06With --style-ref set, refs_for_batch keeps refs[:5] + the anchor (characters, then product, then rooms): a batch with 4 people + band drops its room sheet silently. Plan <=3 named people + 1 room per batch.scripts/cartoon_h3/run.py refs_for_batch
2026-10-06storyboard.py _SYS tells Sonnet to show a tagged speaker 'with their mouth moving' unless the intent is a cutaway, while h3.py says no lip-sync. Dialogue intents must be written as cutaways and storyboard.md grepped for 'mouth' before approval.scripts/cartoon_h3/storyboard.py _SYS; scripts/cartoon_h3/h3.py video_prompt
2026-10-06Cost scale from the 10-04 songs: 1,105 billed H3 s for ~810-830 s of finals (~1.35x incl. regens), ~1.4 canvas renders per batch, 8.4-10.6 s per batch. A 10-min film is ~60 batches, ~$70-98 depending on whether h3-max 768P is $0.05 or $0.08/s.output/dawn-iterate-2026-10-04/video/RUN-LOG.md §4; plan.json of the three 10-04 songs
2026-10-06Photoreal mini-movie is out of scope: Gemini rated Rosabella 11:51 at 9.5/10 difficulty; our avatar gates need one locked character >=8/10 on all shots, and 9-30 Side Effects failed them with one person. REPORT: 'Don't: photoreal mini-movie'.foreplay/14_minimovie.gemini.md; feedback_ai_avatar_broll_lane; REPORT.md §4
2026-10-06reskin.py is song-only: Mode A copies a promoted song run (song/, vo-receipt source=song). A VO/dialogue film re-skin = new slug with identical lines + intents and new style keys; EL per-line cache reuses the takes.scripts/cartoon_h3/reskin.py docstring + copy checks
2026-10-06Audit 10-06: a smartwatch-buzz false solution that fails for minutes undercuts our own touch-gets-through mechanism; default the Ch4 false fix to sound (alarms/phones) and use a wrist buzz only with the 3x-stronger + ramp answer in Ch6.audit_narrative 10-06; product-truth.md motor comparison; survey 09-04
2026-10-06starpop articles: hold rate = first-15-s only at this length; percent-watched is the length signal.~/research/davidaistar/starpop/txt/ (slugs in references/sources.md)
2026-10-06starpop articles: placement gate (Reels safe box y 14-65 %, Feed 4:5 crop) + Advantage+ enhancements-off/live-preview launch check.~/research/davidaistar/starpop/txt/ (slugs in references/sources.md)
2026-10-0610-06: the _SYS 'mouth moving' speaker rule is gone (voice-over staging); gate = python3 -m scripts.cartoon_h3.check_plan --lint output/cartoon-h3/<slug> --pack output/concept-packs/<slug>.md = 0 BLOCK before approval; cutaway wording still needed for --no-llm.scripts/cartoon_h3/storyboard.py, check_plan.py
2026-10-06Built 10-06: run.py --chain-frames (opt-in last-frame start ref on same-scene/env seams, sequential chains, regen cascade), --tier-models, --draft, tiers.md; h3.py prices fixed. Chaining is unproven until a seam-pair proof on Fish's eye.scripts/cartoon_h3/run.py, h3.py, tiers.py (Builder B)
2026-10-06concat.py BUILT 10-06: `python3 -m scripts.cartoon_h3.concat --out <film> <ch1>-final.mp4 <ch2>-final.mp4 [--xfade-frames 6]` (one re-encode, size/fps matched, per-part gain to -14 LUFS, seams in <out>.concat.json). Tested 84.2+89.3 s finals → 173.533 s, -14.01 LUFS. Beds still restart per chapter. ChapterCard comp exists for a Fish-approved test only.scripts/cartoon_h3/concat.py, FormatComps.tsx (Builder C)