Capafy logo Capafy
開始使用
Video Subtitle Translator — Local Kit (Download)

Video Subtitle Translator — Local Kit (Download)

by: LazyAI
Download package that runs on your own computer inside your AI agent (e.g. Claude Code or Codex — its subscription/API costs are yours). Requires Python 3.9+; ffmpeg and faster-whisper for transcription/burn-in (free ~500 MB model on first run). No third-party services or extra fees; your agent translates. Keeps timing, checks reading speed and line length, applies your glossary. Outputs SRT/VTT per language, bilingual subtitles and optional burned-in MP4.
#影片#社群媒體#教育
評分
需要更多評價
已售出
0
使用方式
下載

Video Subtitle Translator

What you're buying: a download package (skill instructions + Python scripts) that runs on your own computer inside your own AI agent — it is not a cloud service.
You need: an AI agent that supports skills (e.g. Claude Code or Codex) — its subscription or API costs are yours; Python 3.9+ (plus ffmpeg and faster-whisper for transcription / burn-in).
Third-party services: none. Transcription runs locally; translation is done by your own AI agent. The Whisper model (~500 MB) downloads free on first run.

One video in, subtitles in many languages out — subtitles people can actually read. Timing stays put, lines never overflow, and viewers can keep up.

What you get

  • 🌏 SRT and VTT for every language — works with YouTube, Vimeo, course platforms and editing software
  • 🔤 Bilingual subtitles: translation and original stacked (original on top or bottom)
  • 🎬 Burned-in video (optional): font picked per language and OS (Microsoft JhengHei, PingFang, Noto Sans CJK, Yu Gothic, Malgun Gothic…), size scaled to frame width — never runs off screen in 16:9 or 9:16

Example

demo_bilingual.webp

A 72-second English talk was translated into Traditional Chinese and Japanese: same cue count and timing as the original, reading speed and line-length checks passed, glossary terms applied. Outputs included a bilingual Chinese + English burned-in video in both landscape and vertical 9:16.

Japanese subtitles, same timing:

demo_ja.webp

Vertical 9:16 for Shorts / Reels — lines never run off screen:

demo_vertical.webp

Try it — copy a prompt

  • "Translate the subtitles of talk.mp4 into Traditional Chinese and Japanese, and burn bilingual Chinese + English into a vertical version."
  • "Here's my SRT — translate it to Spanish, keep 'LazyAI' untranslated, and check reading speed."

Why buy it

One payment, translate as many videos and languages as you like — no per-minute fees.

Who it's for

YouTubers going global, course creators, marketing teams and anyone localising video.

How it works

  1. Source subtitles: local Whisper transcribes the video (99 languages) and splits it into readable cues by sentence, pause and comma — or import an existing SRT/VTT.
  2. Cue-by-cue translation: your agent follows built-in subtitle translation rules; each cue shows how long it stays on screen and how many characters fit.
  3. Automatic checks (fixed and re-checked until there are 0 errors):
    • cue count and numbering must match the source one-to-one
    • glossary: required terms must be translated as specified; locked names and brands must not be translated
    • per-line character limit, max two lines, balanced line breaks (never inside a word, number or katakana term)
    • reading speed: if too fast, it first borrows idle time before/after the cue; only then asks the agent to condense
  4. Export: subtitle files per language, bilingual subtitles, burned-in video.

Language rules built in

Traditional Chinese (Taiwan usage), Simplified Chinese, Japanese, Korean, English, Spanish, French, German, Portuguese, Italian, Indonesian, Vietnamese and Thai have built-in line-length and reading-speed rules; other languages use general rules.

Requirements

  • Python 3.9+ (importing subtitles, translation checks and export need only Python)
  • To transcribe video: pip install faster-whisper (first run downloads a model, small ≈ 500 MB; works offline afterwards)
  • To burn in: ffmpeg (the skill checks your setup and tells you what to install)
  • A normal CPU is enough: the small model transcribes a 72-second video in about 20 s

Limitations (please read)

  • Translation is done by AI — have a fluent speaker review anything important (legal, medical, contracts, brand slogans).
  • Speech recognition can mishear names and numbers; the skill has the agent check the source before translating.
  • Reading speeds follow common web-video practice; streaming-platform specs may be stricter (e.g. 4 Japanese characters per second) and can be adjusted.
  • No dubbing, speaker labels, positioned subtitles or karaoke effects.
  • Only translate videos you own or have the right to translate and republish.

Files

  • SKILL.md — agent workflow
  • scripts/make_source.py — local transcription or SRT/VTT import, source cue splitting
  • scripts/subtool.py — translation template, checks and reflow, SRT/VTT/bilingual export, burn-in
  • scripts/subcommon.py — per-language rules, line-breaking algorithm, subtitle file I/O
  • references/subtitle-rules.md — subtitle translation rules and language notes
  • references/glossary-example.json — glossary example
  • references/example.source.json, example.tr.zh-TW.json — examples

View more