Subtitle Readability Brief
Purpose
Produce a subtitle readability brief before transcription, translation, or ffmpeg burn-in. Competitor skills download, transcribe, or render subtitles — this skill defines how they should read on screen so output isn’t too fast, too long, or inaccessible.
Does not run ffmpeg, Whisper, or download video. Planning only.
When to use
Use when:
- Preparing captions for YouTube, Bilibili, TikTok, course video, or social clips
- Translating or adapting subtitles across languages
- Burning hardsubs (e.g. Chinese via Pillow/ffmpeg workflows)
- User asks about SRT/VTT format, CPS, line length, or SDH
Safety and boundaries
Do not reproduce copyrighted dialogue from proprietary scripts unless the user provides the text.
Do not claim broadcast/legal compliance — note platform-specific limits; user verifies for their jurisdiction.
Sensitive content: flag if verbatim captions could harm privacy; recommend redaction.
Required inputs
- Platform — YouTube, Bilibili, Instagram, broadcast, internal LMS, other.
- Language(s) — source and target if translation.
- Audience — general, children, SDH/deaf/hard-of-hearing, ESL.
- Video type — talking head, fast-cut montage, tutorial screencast, drama.
- Duration — approximate length or “per 60s clip.”
- Style — verbatim, clean verbatim (um removed), summarised.
Default readability targets
Adjust in output; cite platform if known:
| Rule | Typical target |
|---|---|
| Max chars per line | 32–42 (Latin); fewer for CJK if dense |
| Max lines on screen | 2 |
| Max CPS (characters/sec) | 15–17 adult; 12–14 children |
| Min display time | 1.0s (0.83s absolute min for short tags) |
| Gap between cues | 2–4 frames minimum when cutting |
CJK: count characters; Latin: count characters including spaces per platform convention — state which you use.
Workflow
- Confirm platform + language + audience.
- Set numeric limits (table above, tuned).
- Define line-breaking rules (punctuation, clause boundaries, no orphan words).
- Specify SDH needs (speaker IDs,
[music], sound effects) if applicable. - Output brief + sample cue rewrite if user provides a raw line.
Output format
Subtitle readability brief — {platform / language}
| Field | Value |
|---|---|
| Platform | … |
| Languages | … |
| Audience | … |
| Style | verbatim / clean / summary |
| Max chars/line | … |
| Max lines | … |
| Max CPS | … |
| File format | SRT / VTT / ASS / burn-in |
Timing rules
Bullets: min duration, max duration, reading speed exceptions for title cards.
Line-breaking rules
How to split long sentences; forbidden breaks (e.g. between article and noun).
SDH / accessibility
What to tag; speaker labels; when to describe audio.
Translation notes (if applicable)
Register, honorifics, reading speed in target language, expansion factor (~10–20% EN→DE).
QC checklist (before publish)
- No cue > max CPS at display duration
- No more than max lines
- No overlap between cues
- Punctuation consistent
- On-screen text matches safe area (platform margin notes)
Sample cue (optional)
If user gave raw text, show one before → after cue with timestamps placeholder.
1
00:00:01,200 --> 00:00:04,000
First line within limits
Second line if needed
Quality bar
- Numbers are explicit — not “keep lines short.”
- Platform-aware — call out YouTube vs short-form vertical differences.
- Executor-ready — ffmpeg/transcriber skills can apply limits directly.
- Accessibility — SDH section never skipped when audience needs it.
Examples
Good rule: “Max 42 characters per line, max 17 CPS, break at clause commas, never split don't across lines.”
Bad rule: “Make subtitles readable.”
Good SDH note: “Prefix non-visible speaker changes with >> and tag [phone ringing] for off-screen audio.”