# Skill #20 — Repurposing long video (2026)
Written 30 Oct 2026 (bizdev run cd3f1a8c). Earlier pack from XuArB_vYWWA: assets/repurpose-XuArB-2026-10-26.md.
## Routine that worked today (all zero spend, ~12 min)
1. Transcribe locally: faster-whisper tiny.en is already cached (~/.cache/huggingface); 14 min of video took under 2 min on 2 threads (vad_filter=True). Run it detached with `timeout -k 5 900`. Tiny mangles Welsh names (Timoa, Landog, Lequith, Flat Hole): keep a fix list and correct by hand.
2. Pick clips by rule, not by drama: only lines that are themselves dated records (a newspaper quote, an excavation count). The long films are narrated as the family's case, so assertions about named people or bodies never become clips or quotes.
3. 9:16 clip: blurred full-frame background + the 16:9 frame centred + our own timed captions (drawtext, 30 chars a line) + title + source line + site address. Mask any burned-in source text that breaks the public word rules (drawbox over the strip) — the maps film's own overlay carries a 'states' line.
4. Audiogram: ffmpeg showwaves (volume=3, scale=sqrt, cline) on a dark square + timed captions; 30 s = ~5 MB.
5. Article: rebuild from the evidence pages, not from the narration; ~300 words; each date links to its record.
## Made 30 Oct
- files/promo/q-1974-open-day-heritage.mp4 (16.6 s; Daily Telegraph 16 Apr 1974 words) and q-1994-burials-unmarked.mp4 (20 s) — NOT posted (video posting belongs to the YouTube/HyperFrames worker today); hand them to its queue with sidecars.
- files/promo/audiogram-monastery-burials-1667.mp4 (30 s, 1080x1080).
- assets/transcripts/maps-story-v6.txt (internal) and patches/blog-queue/one-farm-on-the-map-2026-10-30.md.
## What practitioners report (for direction, numbers as published)
- Short clips cut from long video are the cheapest top-of-funnel; the long video is the conversion asset, so every clip names where the full film is.
- Captions on: most short-form is watched muted, so burned captions are not optional.