add-captions
GitHub为视频编辑生成符合项目风格的字幕,确保文字简洁易读、位置避开关键画面,并修正转录错误。
Trigger Scenarios
Install
npx skills add Forward-Future/scarlet --skill add-captions -g -y
SKILL.md
Frontmatter
{
"name": "add-captions",
"category": "Picture and graphics",
"description": "Caption the current edit using the project style and short, readable lines."
}
Caption the edit
Caption the speech that is still in the edit, so a viewer with the sound off follows every line without rereading. The brief and the picture decide the wording, placement and style.
Caption what is said, as it is said. Correct what the transcript misheard, above all names, products and numbers, and leave the speaker's grammar alone. A filler the edit kept can stay out of the caption when it carries no meaning.
Break lines where a reader would pause: at the end of a phrase, never between a name and the rest of it, a number and its unit, or an article and its noun. Keep a caption to one or two short lines, roughly 32 to 42 characters each, so at most about 80 characters and 14 words, and on screen long enough to read, around 15 to 17 characters a second, which can mean holding it a little past the words. End a caption with its sentence, and start a new one when the speaker or the shot changes rather than carrying one across it.
Keep captions clear of faces, on-screen text and anything the viewer needs to see. A caption that has to move for one shot can move; the style stays the same throughout.
captions_generate builds the track from the kept words, timeline_propose edits single cues, and frame_contact_sheet shows them on the picture.
Version History
- b6bdb58 Current 2026-11-18 20:45


