Edit plan
Editing is where a good script either keeps its pace or loses it. This skill finds the slow spots in the transcript and tells you exactly where to cut.
Get the transcript
Ask for an SRT or VTT file. Most editors export one (CapCut, Descript, Premiere, DaVinci, YouTube Studio). If there is none, ask for a plain transcript and say the timestamps will be missing, so cuts will be by line, not by time.
Read the pace
Tool subtitles_analyze with the file as text (CLI: python scripts/sr.py subtitles --file talk.srt). It reports:
- words per minute overall (a rough guide: about 150 is a relaxed conversational pace, and fast, punchy short-form often runs higher; compare with the person's own best videos, not with a rule)
- cut list: gaps longer than the pause threshold (default 0.7 s), with timestamps and how many seconds you would save
- filler words inside cues (um, uh, you know, okay so) with timestamps, to trim by ear, because an SRT has no word-level timing
Apply the cut list in the editor, top to bottom. Leave a short breath where a pause carries meaning (after a punchline or a reveal).
Caption file for vertical video
Tool subtitles_rechunk (CLI python scripts/sr.py rechunk --file talk.srt > talk_vertical.srt). It re-times the words into short balanced lines of about four words and 28 characters, free to cross the original cue edges, with no one-word orphans. Import the result into the editor and style it once.
Style tips: one line, high contrast, a highlight on the key word, kept away from the platform's own buttons at the bottom and right edge.
Pacing
Craft rules of thumb, not laws: change something on screen every few seconds (cut, zoom, text, b-roll). Use sound to mark beats. Cut the first word of a sentence if it is "so" or "and". Put a visual change inside every beat that the beat sheet flagged as long on one line or without a number, name or turn (sr-script, tool script_beats).
Plan b-roll and on-screen text
Write a table:
Time | Line | On screen | Note
The on-screen text should add to the spoken line. Do not subtitle the same words twice in different places.
Export
Vertical video is 9:16, 1080 by 1920. Ask the person to check the platform's current guidance in the app for frame rate, bitrate and file size. Keep a clean master without burned-in captions for repurposing.
Gate
human: the person edits. When they return with the cut, ask them to export retention after publishing and run sr-audit.
Do not
- Remove every pause. Silence is part of rhythm.
- Add background music so loud that it fights the voice.
- Use music you do not have rights to (
sr-policy-check).