How to translate a video into another language

You have a video that works in one language and an audience that does not speak it. There are three levels of answer, they cost very different amounts, and the cheapest one is not the weakest.

Level 1 — Subtitles

Burn or generate translated captions and leave the audio alone.

This is the cheapest option, and on social it is frequently the best one: most feed video is watched muted, so the subtitle is the audio for a large share of the audience regardless of language. It also never puts a wrong mouth on screen, which is the failure mode people notice most.

Its limits are real too. Reading competes with watching, so anything visually dense — a screen recording, a demo, a diagram — is worse with subtitles than with a dub. And a subtitle cannot help a viewer who is driving, cooking or looking away.

Level 2 — Dubbed audio

Replace the voice track with the same content spoken in the target language, and leave the picture as it is.

This works well whenever the speaker is not on camera: voiceover-led explainers, b-roll montages, product tours, faceless channels. Generate the new narration as a voiceover and mix it in place of the original. Nothing about the picture has to change, because no mouth is visible to disagree with the audio.

Level 3 — Dub with the mouth re-synced

When the speaker is on camera, laying a translated track over the picture reads as dubbed, and viewers disengage from it. The video translator does the whole job in one pass: the speech is translated, spoken back in the target language, and the mouth is rebuilt around the new audio, so it still reads as the same performance.

Practical notes:

• Thirteen-plus languages, picked from a list rather than typed, so a typo cannot cost you a run.
• Engines differ on the voice — some carry the original speaker's voice into the new language, others use a matched voice. The picker says which is which, and delivery and timing follow the original take either way.
• Priced by clip length, quoted before you run. One engine bills per started minute, which matters for the ordering below.
• Re-voicing without translating is the same machinery: lip sync takes any audio you hand it, in any language, and syncs the mouth to it.

Lock the edit first

This is the one rule that costs money when broken. Dubbing consumes the clip as it exists, so any cut you make afterwards invalidates the dub for that clip and you pay to run it again.

Finish the edit. Then dub. Then caption, against the new audio, not the old.

Related: when a run is billed per started minute, twelve five-second clips cost twelve minutes. Dub the continuous talking clip in one run rather than piece by piece, and split it afterwards if the edit needs the pieces separated.

Where translated video goes wrong

Length drift. Spanish, French, German and Portuguese all run longer than the same sentence in English — commonly 15–30% longer. If the picture was cut to the rhythm of the original narration, the translation will not fit it. Symptoms: dialogue running past the cut, or a lip-synced mouth still moving after the shot changes. Give the picture room where you can, and expect to adjust a few clips rather than none.

On-screen text. Titles, lower thirds and captions burned into the picture stay in the original language and quietly announce that this is a translation of something else. Text you added on the timeline is quick to edit; text baked into source footage needs to be covered, re-generated, or cropped out.

Names and terms. Product names, brand terms and jargon get translated when they should not be. Read the transcript before you commit a run, and expect to correct the same handful of words in every language.

Numbers, dates and units. A localised video that quotes prices in the wrong currency and dates in the wrong order reads as machine output, because it is.

Nobody checked it. Get a speaker of the target language to watch it once, end to end. Machine translation is fluent and confidently wrong in ways that are invisible to someone who does not speak the language — this is the single highest-value ten minutes in the process.

Then re-caption

Run auto-captions against the dubbed audio, not the original. You want the subtitles to match the words actually being spoken, and the muted-autoplay audience to get the translated version too. Keep lines short — two or three words at a time reads far better than a wrapping sentence, in every language.

Finally, export each language as its own file. One video per language, per platform, with the captions to match.

The order that works

1. Choose the level: subtitles for reach and cost, dub for voiceover-led material, dub with lip sync when a face is talking.
2. Lock the edit before anything is dubbed.
3. Dub the whole talking clip in one run, especially when billing is per started minute.
4. Fix the on-screen text, names, numbers and units.
5. Absorb the length drift — translated speech runs longer than English.
6. Caption against the new audio.
7. Have a speaker of the language watch it once before it ships.