To translate video captions on iPhone, generate captions first, then translate the track in place with a tool that re-flows each line onto its original timing. The trap is sync: a naive translation swaps the text without re-timing it, so word-pop animation highlights the wrong words. Zella translates into 19 languages entirely on-device via Apple's Translation framework, needs iOS 18 or later, and never sends the transcript anywhere.
To translate captions on iPhone, generate captions first, then translate the caption track in place with a tool that re-flows each line onto its original timing. Translating the text alone is the easy half; keeping it in sync is what makes the result watchable.
What iOS gives you, and where it stops
Apple ships more translation than people realise, and none of it touches a video file.
- The Translate app handles typed text, live conversation and text through the camera. It has no notion of a video or a timeline.
- Live Captions (Settings > Accessibility > Live Captions) transcribes audio on-device while you watch. It is a viewing aid drawn on your screen, in your language: it produces no editable track and never reaches an export.
- Photos has no captions and no subtitles.
- iMovie has no automatic captions, so there is nothing to translate. Its titles are typed by hand, one at a time.
- YouTube can auto-translate its own captions for viewers, but that is YouTube playback only. Nothing carries to TikTok, Reels, or a file you hand a client.
So the working order is fixed: you need a caption track with real word timings first, and then a translator that understands those timings.

Transcription of this kind runs through Apple's Speech framework, which supports on-device recognition, so the audio does not have to leave your device.
Captions are an accessibility requirement before they are a retention tactic: the W3C's guidance on captions for prerecorded media is the bar most teams are measured against.
Why translated captions usually break
A caption track isn't a block of text - it's a sequence of timed words. Word-pop and karaoke styles highlight each word as it's spoken, which means every word carries its own start and end.
Translate line by line and four things go wrong:
- Word counts change. "Thanks for watching" is three words; its Spanish equivalent may be four. Highlighting now drifts.
- Word order moves the keyword. German pushes the verb to the end of a subordinate clause; Japanese puts it at the end of the sentence. If your emphasis was on English word three, word three of the translation is usually not the same idea.
- Emphasis stays on an index, not on a meaning. A naive translator keeps the highlight on whatever position it had, so the wrong word pops.
- Line breaks stop matching the breath. Longer languages overflow the window they were sized for.
The fix is to spread the translated line across the original window and re-pick the words to emphasise in the target language.
The languages that carry extra traps
- Right-to-left. Arabic reverses reading order. Check caption alignment, and check any arrow, underline or callout that was pointing at a specific word.
- No word spaces. Japanese, Chinese and Thai do not separate words with spaces, so a highlighter that splits on spaces will pop the entire line at once. Segmentation has to be character-aware.
- Expansion. German and Spanish commonly run 20-30% longer than English. A line that fitted one row now wraps to two, which on a 9:16 frame can push it under the platform's caption block.
- Line length. Subtitle practice caps a line at around 42 characters and two lines at a time. It is a good ceiling on a phone too, where the text is competing with the video for a small frame.
Doing it on-device matters
Cloud translation means your transcript - sometimes the whole video - goes to a third-party server. For client work, anything under NDA, or personal footage, that's a real consideration. On-device translation has no network path at all, and works on a plane.
Step by step
- Generate captions. Translation needs a caption track to work from.
- Fix names and jargon first. Correct wrongly-transcribed names, brands and technical terms before translating - errors get faithfully translated too, and a mangled name is worse in two languages than in one.
- Download the language model once. iOS keeps offline translation models per language. On iOS 18 they live at Settings > Apps > Translate > Downloaded Languages (on older layouts, Settings > Translate), where On-Device Mode is also switched on. Do this on Wi-Fi before you travel.
- Pick the target language and translate.
- Check the longest lines. Skim for lines that now crowd the frame and split them.
- Re-read the emphasis. The word you wanted to punch in English is rarely in the same place.
- Export one cut per language, and write the post's own caption, title and hashtags in that language too. Otherwise you have translated the video for an audience the platform will never show it to.
How Zella does it
Zella for iPhone & iPad translates captions into 19 languages - Spanish, Hindi, Portuguese (Brazil), French, German, Arabic, Japanese, Indonesian, Chinese (Simplified), Korean, Italian, Russian, Turkish, Vietnamese, Thai, Dutch, Polish, Ukrainian and English (UK) - entirely on-device via Apple's Translation framework, with no network path.
Each translated line is re-flowed onto its original on-screen window, so word-pop and karaoke keep animating at the right moments, and the strongest words are re-emphasised in the new language rather than left on the English ones.
Two honest caveats. It requires iOS 18 or later, and the first use of a language may need a one-time offline model download from Settings. And it is an iPhone and iPad feature: the Mac app does not have caption translation. Translation is free, like the rest of the captions toolset. The Mac app ships on the Mac App Store today; Zella for iPhone and iPad is coming to the App Store. See the captions docs.
FAQ
Do I need internet to translate captions? Not in Zella - translation runs on-device. iOS may need a one-time download of a language's offline model, after which it works in airplane mode.
Can I do this on the Mac app? No. On-device caption translation is an iPhone and iPad feature, because it uses Apple's Translation framework on iOS 18 and later.
Will the translation be perfect? No machine translation is. It's reliably good for conversational speech and weaker on idioms, jokes and industry jargon - read it through before publishing, especially for anything commercial.
Can I show two languages at once? You can, but it eats the frame and neither reads well. One language per export, with the language named in the post's own caption, performs better.
Does translating change my video's timing? No. Only the caption text changes; the video, cuts and audio are untouched.
Make your next video with Zella.
Record, edit and ship on Mac, iPhone and iPad - local, private, free to start.
RELATED