사용 방법
- 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
- 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
- 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/appariciojunior/video-editor-skills --skill captions시작 전 확인
- 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
- 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.
Captions
Goal: people watching on mute can follow; the captions are just there, never in the way.
Source
The verbatim transcript mapped onto the current timeline (follow the creator's trims), not Resolve's own auto-captions (a second, less accurate transcription). Drop words on B-roll (radio lyrics, background talk).
Clean, not verbatim
Captions show what they mean, written correctly:
- Stutters: "pro-product" → "product", "da-da," → "da,".
- Cut-off words: "es--", "A--" → gone.
- Filler sounds: drawn-out "uuum", "hããã", "ééé" → gone. Careful with real words that look like fillers ("um" = "a" in Portuguese, "é" = "is").
- Self-corrections: "get a, an appointment" → "get an appointment".
- Spoken short forms → correct written forms, if the creator asks. Brazilian Portuguese example: cê/cês → você/vocês, tá → está, tô → estou, tava → estava, tamo → estamos, vamo → vamos, pra/pro → para/para o, né → não é, "o que que" → "o que", "por que que" → "por que", "tu consegue" → "você consegue"; "tão" → "estão" only before a verb in -ndo ("tão bonito" stays). Keep expressions and swear words (fix grammar, not voice). Keep capital letters at sentence starts when you shorten.
- Names and spellings checked on the web (transcripts garble product names), and the creator's own name exactly as they write it.
- Don't reveal what the creator wants kept secret (a place, a company) even if they say it.
Keep two lists: global rules (word map + phrases) in config, per-video fixes in caption_fixes.json. Apply phrase fixes after the word map.
Lines
- One line, max 42 characters; a sentence's last word may stretch it to ~50 rather than sit alone.
- New cue at a pause ≥ 0.6 s, at sentence ends (if the line is already substantial), or after 3.2 s.
- Min 0.9 s on screen, end 0.25 s after the last word, never overlapping the next.
- A lone small word after a pause ("que", "vai", "not") joins the next line. Single-word cues are fine only when they stand alone ("Calm down.", "Bye.").
- First cue starts at 00:00:00,000 (Resolve's import shifts everything relative to the first cue).
- Write a new file name per version (
captions_v3.srt).
Read every cue
Print every line and actually read it before handing it over. Grep for leftover short forms, double punctuation (",,"), lowercase after a full stop, words lost by your own filters. Fix rules, regenerate, read again.
Into Resolve
The creator imports with File > Import > Subtitle (the API places subtitles after the timeline end; a workaround crashed Resolve). Delete the old subtitle clips but keep the track so its style stays.
Style
Record the creator's style once (font, size, position, shadow) from their timeline and save it in config as the house style, e.g. a light sans at size 68, centred ~11% up from the bottom, soft drop shadow. Overlays must stay out of that band (see motion-overlays).