How to translate video subtitles with AI

Translated captions are the cheapest way to reach a new audience. AI can now produce a usable first draft in seconds — but only if you understand where machine translation is strong, where it slips, and how to check the result without speaking the language fluently.

Updated 1 October 20269 min readIntermediate

In short: transcribe in the original language first and fix it; then translate. Keep sentences short when you speak, and have a native speaker skim anything important — names, numbers and jokes are where machines fail.

Why translate at all

Around three quarters of the world's internet users don't speak English as a first language. A cooking video in Spanish, a tech review in Hindi or a travel vlog in English can each find a second audience that its creator never reached — and platforms recommend videos to viewers whose language matches the captions. Dubbing is expensive; subtitles are nearly free.

How the translation works

VDOAPP translates in up to two steps, all on your device:

  1. Speech to English. Whisper, the speech model used for captions, can translate as it listens: it hears Bengali or Japanese and writes English directly. This is often better than transcribing and then translating, because it works from the sound rather than from a possibly imperfect transcript.
  2. English to your target. For other target languages, each English phrase is passed through a compact translation model (the open Opus-MT family, one small model per language pair). Whole phrases are translated before they're split into caption lines, so grammar that spans a line break survives.

If the video is already in the target language — you asked for Spanish captions of a Spanish video — VDOAPP notices and simply transcribes it.

Which languages to target

The Show captions in menu offers English, Spanish, French, German, Italian, Hindi, Arabic, Chinese, Japanese, Russian, Indonesian and Vietnamese. To choose:

  • Look at your analytics. YouTube Studio and most platforms show viewers by country. A cluster of viewers in Indonesia or Mexico is a signal.
  • Match your topic to a market. Cricket content → Hindi; football → Spanish; anime and games → Japanese.
  • Start with one. One well-checked translation is worth more than five unchecked ones.

Step by step

  1. Caption in the original language first and proofread. This tells you the speech was heard correctly.
  2. Export .srt of the original — useful to keep, and to upload as a second caption track.
  3. Choose the target in Show captions in, keep the spoken Language set correctly, and press Generate captions. The new captions replace the previous ones on the timeline.
  4. Review (see below), then export the video or the translated .srt.

Checking a translation when you don't speak the language

  • Numbers, names and units must match the original exactly. These you can check in any script.
  • Line lengths — German and Russian run 20–30% longer than English; Chinese and Japanese much shorter. Very long lines may need splitting.
  • Back-translate a sample. Paste a few lines into any translator going the other way. If the meaning comes back mangled, rewrite the original sentence more simply and try again.
  • Ask one native speaker to watch the first minute. Tone problems (too formal, too casual) show up quickly.

Common pitfalls

ProblemWhyFix
Jokes and idioms fall flatTranslated word by wordRewrite the caption freely, or add a short explanation
Wrong formality ("tu" vs "vous")English doesn't mark itPick one and edit consistently
Product names translatedModel treats them as wordsRestore them by hand
Right-to-left text looks oddArabic mixed with numbers or Latin wordsKeep Latin words at the line ends where possible

Burned-in captions or an .srt file?

Burned-in captions (part of the exported picture) show everywhere, including on platforms without caption support, but viewers can't turn them off and you can only have one language. .srt files uploaded to YouTube or Vimeo let viewers pick a language and are indexed by search. The best approach for a multilingual audience: burn in the main language and upload translations as .srt tracks.

Frequently asked questions

Is AI translation good enough to publish?

For casual, informative videos, usually yes after a quick check. For legal, medical or brand-critical content, have a professional review it.

Can I translate into a language not in the list?

Not inside the editor yet. Export the English .srt and translate it with another tool, then import it with Import .srt.

Does translation need an internet connection?

Only to download each model the first time. After that, translation runs offline on your device.