CaptionSync Pro

CaptionSync Pro

Display name: CaptionSync Pro Technical slug: subtitle-sync-pro Seller: Apex Studio LLC Category: Translation and Localization Price: $9.99 one time, paid skill License: Apex Studio LLC Paid Skill License (see LICENSE). No redistribution, no resale.

One line

Translate subtitles into 44 languages, then retime every cue so viewers can actually read them.

Short description for the card

Machine translation gets the words right and the timing wrong. CaptionSync Pro translates your transcript or SRT/VTT file, then recomputes each cue from a per language reading speed budget, splits what cannot fit, protects your glossary terms, and hands you a QC report that lists every cue still needing a human eye.

Why it matters

German often runs longer than English. Japanese is read by character, not by word. Arabic reads right to left. Paste translated text onto the original timing and viewers get subtitles that flash past or overflow the screen. This skill fixes the part machine translation ignores.

What it does

• Takes a plain text transcript, an SRT file, or a VTT file
• Translates into 44 languages, with Google Cloud Translation as the default provider and your own key
• Retimes every cue for its language: needed time is characters divided by that language's reading speed, never below the minimum, and cues are only ever extended, never shortened
• Keeps start times locked to your audio when the file has real timing
• Splits cues that cannot fit on two lines or run over the maximum duration, at a clause or word boundary, and shares the time between the pieces
• Keeps two speaker dialogue one speaker per line, and splits it between speakers when it cannot fit
• Protects glossary terms such as product names and brand names from being translated, split, or altered, and checks each language afterward
• Writes SRT and VTT for every language, plus a QC report

Retime only mode

Already have subtitles that read too fast? Set source and target to the same language. Nothing is translated and no API key is needed.

What you get in the QC report

For every language: cues over the reading speed budget, cues under the minimum duration, auto splits, extensions, line overflow, glossary violations, runtime change, and a table of the exact cues that need manual review. Nothing is hidden. If the next cue starts too soon to extend a cue, it is flagged instead of silently squeezed.

Proof from the bundled sample

Episode 12, 14 cues, translated with supplied files:

• Spanish: 14 cues in, 16 out, 0 over speed, 0 review items
• German: 14 cues in, 17 out, 3 flagged for review, each listed in the report
• Japanese: 14 cues in, 17 out, 0 over speed, 0 review items
• Arabic: 14 cues in, 14 out, 4 review items listed

The remaining German cues are back to back with the next cue, so their start times cannot move. The report says so rather than pretending they are fixed.

What is in the package

• SKILL.md: the instructions your agent follows
• Eight Python scripts plus a runner, Python 3.9 or newer, standard library only, nothing to install
• Reading speed budgets and language data for 44 languages, editable in two small tables
• A sample transcript, glossary, hand written translations and golden outputs, and a test suite of 129 checks, all passing
• Hero image: captionsync-pro-hero.webp

Requirements

• An agent that supports skills and can run Python
• For translation with the default provider: a Google Cloud Translation API key, supplied as an environment variable. Google bills you per character on your own account. The skill never prints or stores your key.
• No key needed for retime only runs, or if you supply translations as files.

Honest limits

• Translations are machine output. Have a native speaker check them before publishing.
• Without an audio file, plain text transcripts get estimated timing at 150 words per minute. Review it against the audio. There is no forced alignment and no transcription.
• Cues are translated one at a time, so a sentence that runs across several cues can read less naturally.
• Chinese, Japanese, and Korean line breaks use no dictionary. Katakana words and numbers stay whole; other breaks can still land inside a word.
• Right to left text is checked, not reshaped. Mixed direction lines are flagged.
• The default reading speeds are sensible starting points, not a platform standard. Pass your own with a per language override when needed.
• Not for dubbing, burning subtitles into video, or transcribing audio.

Example prompt

Translate episode12.srt into Spanish, German, Japanese, and Arabic. Keep MiraDiagnostics and Hands In untouched. Give me SRT and VTT, and tell me which cues need review.

Try it first

Start with retime only on a file you already have. That shows the timing fix with no API key and no translation cost.