Skip to content

konradre/video-production-skills / skills/spot-audio-assembly

spot-audio-assembly

Verified

Assembles the SOUND of a generated-video piece: voice-over takes (one voice and model per role), dubs for refused lines in the character's cloned voice, off-screen lines, sfx with the syllable map confirmed before the cut, music cues cut to markers with ducks, room tone, captions, the stem, and the read of the delivered audio. Use when a line must be voiced or re-voiced, a line was refused, sfx or music must be sourced or cut, captions must be built, a take's audio carries an artifact, or the delivered audio must be verified. Triggers — "generate the VO", "same voice as before", "dub the line", "the off-screen voice", "the sting", "the syllable map", "the music is too loud", "duck the music", "room tone", "captions", "is the vo on time", "musical artifact at the start". Not for placing lines on the timeline or the beat gate — use video-edit-edl. Not for the master render or the delivery encode — use video-finish-qc. Not for LUFS mastering of a finished mix — use mastering-audio.

FileKindSize
SKILL.mdskill md22.0 KB
references/CAPTIONS.mdmarkdown5.4 KB
references/MIX-AND-QC.mdmarkdown13.4 KB
references/SFX-AND-MUSIC.mdmarkdown13.2 KB
references/VO-PIPELINE.mdmarkdown6.6 KB
scripts/audio_head_scan.pyexecutablescript3.9 KB
scripts/beat_grid.pyexecutablescript30.1 KB
scripts/build_captions.pyexecutablescript8.8 KB
scripts/build_vo_stem.pyexecutablescript5.7 KB
scripts/gen_vo.pyexecutablescript7.0 KB
scripts/music_loop.pyscript7.6 KB
scripts/music_take_read.pyscript28.2 KB
scripts/qc_vo_placement.pyexecutablescript4.2 KB
scripts/roomtone_synth.pyexecutablescript3.4 KB
scripts/sfx_onsets.pyexecutablescript7.2 KB
scripts/sound_table.pyexecutablescript33.5 KB
scripts/vo_word_times.pyexecutablescript6.4 KB
scripts/voice_ab_reel.pyscript3.9 KB

references/CAPTIONS.md

View on GitHub
# Captions — the campaign standard

One style for every clip of a campaign, fixed in the EDL's `caption_style` and rendered by
`build_captions.py` as RGBA cards at DELIVERY resolution, which the finisher overlays AFTER the downscale
so the type stays crisp.

## The house default

| rule | value |
|---|---|
| font | one display face (style `A`), white, black stroke, soft shadow, centred at 70 % height, max 86 % width |
| case | ALL CAPS |
| punctuation | **NEVER** — no periods, commas, question marks, ellipses; contractions keep the apostrophe; hyphenated compounds split into two words |
| highlight | the spoken word in the brand colour (a colour taken from the wordmark) |
| whose lines | the narrator only; never a character's dialogue, never an O.S. line |
| cards | word ranges per line; three-word cards for a fast disclaimer; a quip's card follows its VO variant exactly |
| timing | show 0.05 s before the first word, hold 0.5 s after the last; a card ends **one frame before the next card's lead**; nothing over the end card |

A card's line breaks come from the font's measured advance — `build_captions.py` lays each line out with the TTF's
own `textlength` against the card width — never from a character count, which mixed scripts and wide capitals defeat.

## A campaign's own style — set by measurement

When the client's reference carries captions, measure them on a frame at the reference's own width — cap height,
line pitch, block centre, stroke, shadow, case, emphasis — and set the style from those numbers, never from an
adjective ("subdued", "UGC-like"). Punctuation stays banned unless the operator lifts it; a reference that shows a
comma does not lift the rule.

| key | reads | default |
|---|---|---|
| `font` | `A` Bangers · `B` Inter (variable) · `C` Nunito (variable) · or a font file name, under the fonts dir | `A` |
| `weight` | a variable font's named instance (`Black`, `SemiBold`) | `Black` on `B` |
| `size_h` · `stroke_h` · `shadow_h` | fractions of the canvas height | 0.058 · 0.0055 · 0.004 |
| `shadow_alpha` | 0–255 | 150 |
| `line_h` | line pitch as a multiple of the size | 1.12 |
| `y` · `max_w` | block centre (fraction of height) · card width (fraction of width) | 0.70 · 0.86 |
| `highlight` | `#rrggbb`, or `none` — one plain layer per card, no per-word emphasis | `#ffe01b` |
| `upper` | ALL CAPS | `true` |
| `lead` · `hold` | seconds before the first word · after the last | 0.05 · 0.5 |

**An unknown key stops the build.** A style block whose key names matched nothing the builder read once rendered the
defaults for a whole job and looked deliberate. Keys starting with `_` are notes.

**Placement is checked against OUR frames, never copied from the reference.** A block centre measured on a client's
reference sat across the faces of a two-shot whose subjects sit higher in frame than the reference's: read the caption
band against every event's frames at the cards' display times before the render ships.

## The words

The display text is the script's (`audio.vo[line].text`); the transcript supplies timing only. A transcript's
capitalisation is not the client's copy — a mid-sentence word came back capitalised, invisible under ALL CAPS and
wrong the moment the case changed — so script word *i* pairs with time *i* only when the two counts agree, and the
builder stops otherwise (re-run that line's word times with `--only`).

## Word times

`vo_word_times.py` writes `{LINE: [[word, start, end]…]}` relative to each VO file; timeline time is
`vo[line].at + word time`. Local faster-whisper is free and merges hyphen fragments; Scribe bills per
minute. A swapped take re-runs that line ONLY (`--only`) so hand patches on the others survive. Whisper
mis-times onsets in noisy sections — patch by ear against the placement check, never by re-running.

## Card boundaries — half-open, never `between()`

The cards are PNG layers composited by `video-finish-qc/scripts/finish_spot.py`, and the expression that
switches them is load-bearing. **`enable='between(t,a,b)'` is inclusive at BOTH ends**, so two layers that
share a boundary are both enabled on the frame that lands exactly on it. Every word-highlight layer inside a
card shares one by construction: `build_captions.py` hands layer *k* the `t1` that is layer *k+1*'s `t0`.
Card-to-card the builder already leaves a frame of air (`nxt - LEAD - 1/24`), so a card-to-card collision
cannot arise; the exposure is the highlight layers alone. No delivered spot is known to have carried the
defect — what follows is the mechanism reproduced in a scratch render, not an incident.

Measured 2026-09-17 in a scratch render, 24 fps, boundary at t = 1.0: with `between()` one frame composited
both layers; with `enable='gte(t,a)*lt(t,b)'` zero frames showed both **and** zero showed neither. The
half-open form is the fix. **An epsilon trim is not** — subtracting 2 ms from the earlier layer trades a
collision that needs exact equality for an uncovered window 2 ms wide, and at 24 fps a frame lands in that
window about one time in twenty. A doubled frame is a smudge; a blank frame is a missing caption.

A designed card that carries a **colour emoji** is rendered by PIL with `embedded_color=True`; ffmpeg's
`drawtext` cannot draw one at all, which is the reason these are PNG overlays rather than a text filter.

## Alternates

An alternate quip or an alternate cut is its own EDL with its own captions; a caption never carries a
line the deliverable does not voice.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 196kUpdated Claude CodeCodexVerifiedInstall