Workflow guide
Subtitles that read natively in every language you ship
Translated captions fail in predictable ways: text that overflows the card, jokes that land flat, and timing tuned for one language forced onto another. Translation is a second edit, not a find-and-replace pass. This guide shows how to carry timing, meaning, and readability across languages.
From the Mellius product team · Updated September 28, 2026
Conspiracy theorists, you may hate this
AI Director
›
Just type what you need
- ›Remove all silent parts
- ›Add bold kinetic captions
- ›Only have Bill Ackman speaking
- ›Cut to highlights only
- ›Add cinematic transitions
- ›Match cuts to the beat
The short answer
Export your original captions as SRT, machine-translate the file, then have a native speaker fix idioms, names, and line breaks so each card stays under 42 characters and one to six seconds. Re-import, spot-check sync against speech, and export sidecar SRT for YouTube plus burned-in versions for vertical feeds.
Timing survives translation
Every translated card still appears with its line and clears before the next idea starts.
Meaning stays local
Idioms, humor, and names read naturally to a native speaker, not like a dictionary lookup.
One master, many exports
A single timed master produces sidecar files and burned-in renders for each language and platform.
Translate the file, not the timeline
Always translate from an exported SRT or VTT rather than retyping on the timeline, because the file carries the timing your first language already earned. Run machine translation for the rough pass, then hand it to a native speaker with the video, not just the text — translators who can hear tone and see the picture catch sarcasm, formality levels, and gendered address that raw text hides. Lock proper nouns, brand terms, and recurring phrases in a small glossary first so episode ten matches episode one.
Expect text expansion: English to German or Spanish typically grows 15 to 30 percent, which breaks line lengths tuned for the original. Split expanded lines at clause boundaries and keep 32 to 42 characters per line with at most two lines per card. In Mellius, translated captions land as editable timeline objects, so a fix is a sentence away and undo restores the previous version.
- Export SRT/VTT from the finished source captions before translating.
- Give translators the video plus a glossary of names and brand terms.
- Plan for 15–30% text expansion into German, Spanish, and French.
- Keep one idea per card; split at clauses, never mid-phrase.
Fix timing, breaks, and cultural misses
Re-import the translation and watch it against the audio at full speed, because a card timed for a three-second English line may need five seconds in another language. Where speech runs faster than comfortable reading speed — roughly 160 to 180 words per minute — merge short cards or simplify wording rather than flashing text viewers cannot finish. Keep each card on screen one to six seconds with at least two frames of clearance across shot cuts.
The classic pitfalls are cultural, not technical: date and number formats, humor that references the source culture, on-screen text that also needs translating, and reading direction for right-to-left languages like Arabic and Hebrew. Burn a test render of the densest minute and watch it on a phone before committing to the full export. Common mistake: trusting platform auto-translate for anything beyond a draft — it preserves timing but mangles names, homophones, and tone.
- Re-time cards to the dubbed or original audio, whichever viewers hear.
- Localize dates, units, humor, and visible on-screen text.
- Test the densest minute on a phone before full export.
- Mirror layout and check fonts for right-to-left languages.
Export and QA one language at a time
YouTube, Facebook, and LinkedIn accept multiple sidecar caption tracks, so upload one SRT or VTT per language with clean locale names rather than merging everything into a single file. TikTok, Reels, and Shorts need burned-in translations, which means a separate vertical render per language with text positioned above platform UI overlays. Keep a folder per locale containing the SRT, the burned-in master, and the glossary version used.
Run a final QA pass per language with three checks: sync against speech, spelling of every proper noun, and overlap with graphics or faces. If a fix touches timing, re-export that language's sidecar from the corrected timeline rather than hand-editing timestamps. In Mellius, each language version stays reversible on the timeline, so late client notes do not mean rebuilding the project from scratch.
Compare the workflows
| Method | Best fit | Trade-off |
|---|---|---|
| Platform auto-translate | Draft subtitles and personal viewing | Unreviewed errors in names, tone, and timing |
| Machine translation plus native review | Most creator and marketing work | Needs a reviewer who sees the video |
| Professional transcreation | Drama, comedy, and brand-critical releases | Slower and priced per minute |
A practical sequence
Put it into practice
- 01
Export and machine-translate
Export the finished source captions as SRT and run a machine pass per target language.
- 02
Native review with picture
Have a native speaker fix idioms, names, and line breaks while watching the video.
- 03
Re-import, spot-check, export
Re-time cards to speech, then ship sidecar SRT and burned-in renders per platform.
Frequently asked questions
Should translated captions be burned in or uploaded as SRT?
Upload sidecar SRT or VTT on YouTube, Facebook, and LinkedIn so each language stays toggleable and crisp. Burn translations in for TikTok, Reels, and Shorts, where sidecars are unsupported and cropping can bury lower-third text. Keep a per-locale folder with both so future fixes touch one source.
Why do translated subtitles overflow or flash by too fast?
Most target languages expand 15 to 30 percent over English, so cards tuned for the original no longer fit. Split expanded lines at clause boundaries, merge cards where speech outruns reading speed, and re-time durations to the audio viewers actually hear. Always re-check the densest minute on a phone.
Can I trust YouTube auto-translate for my videos?
Auto-translate is a serviceable draft for casual viewing but a risky master for published work. It preserves rough timing while mangling proper nouns, homophones, humor, and formality levels. For anything public-facing, run a native-speaker review pass with the video before publishing.
Continue reading
Primary sources
Product capabilities change. These official references support the comparisons above.
Describe the edit. Keep the timeline.
Mellius turns conversational direction into cuts, captions, graphics, motion, color, and sound you can still inspect and change.
Start editing free