Translate the video into another language

Translate the spoken track, then caption the new language. Translate a video into another language is dub_video with translate_audio_only. It is lighter than a cloned dub. Auto-captions will transcribe whatever is now on the file. If you caption first in English, then translate, the burn still says English while the speaker says French.

This is audio translation plus a new caption pass. It is not “dub without a new mouth” (that page is whether visemes must match). It is not the order-of-operations essay titled dub-first-then-caption.

Translate the track, then caption what is now spoken

add_veed_captions transcribes. It does not translate. The language code you pass must match the language actually spoken after the job. 165 codes exist. fr-FR on an English file will not magically subtitle French; it will try to hear French in English audio and miss.

So the sequence is:

1. Picture lock. Do not translate a take you will recut.
2. Translate or dub the soundtrack onto a Versely-hosted URL.
3. Listen. Confirm the offer, the SKU, the legal line.
4. Caption the translated file in the new language.
5. Overlay anything that must stay exact and was never spoken: a URL, a mg amount.

Source constraint: the media must already live on Versely storage — a prior generate, upload, or output URL. YouTube links are not a dub source. Download, upload, then run.

The job is async. dub_video returns a project_id. Poll get_dub_status (queued → dubbing → completed or failed). Do not treat the same chat turn as the finished MP4. list_dubs browses recent projects.

source_lang is optional if auto-detect is acceptable. Set it when the plate is mixed, noisy, or a language the detector will confuse. target_lang is not optional in spirit: pick the market, not “European.”

Cost scales with length. estimate_cost before a 20-minute course. check_credits if the balance is tight. There is no flat number in the tool spec to quote as a Versely price.

translate_audio_only is not a cloned dub

Same tool, different depth.

translate_audio_only. Spoken content into the target language without the full voice-cloned, engine-choice pipeline. Lighter. Good for internal review, a first pass for a local team, a course draft, a clip that must be understood before you spend on a performance. It will not reposition picture. It will not retarget a mouth.

Full dub. Default engine elevenlabs: clones delivery, audio or video, trim supported, up to 30 minutes, no mouth retarget. Engine heygen: video-only, lip-sync to the new language, 8-minute cap, fewer languages, no start/end trim. Use full dub when the market has to believe it is the same person speaking Spanish. Use translate-audio-only when they have to understand the sentence.

Mouth-is-the-product still matters. A talking-head close-up in a new language with old visemes will look dubbed because it is. If that is unacceptable, you owe heygen or a dedicated lipsync after a new wav — a different spend. If the plate is pack, hands, pour, a small PiP, translate-audio-only plus a caption is enough.

NY GBL § 396-b (in force 9 June 2026) exempts AI used solely for language translation of a human performer from the synthetic-performer ad disclosure. That is a statute, not a Versely price. It does not exempt photoreal AI picture. Article 50 (2 August 2026) still wants a deployer label at first exposure when the image would pass as real. Translation of the track does not wash the picture’s disclosure.

Do not run translate-audio-only on a silent row and hope. There is nothing to translate. Generate a VO in the target language instead and replace.

Burn captions in the new language

After the translated audio is on the file:

1. Watch once. Note names, numbers, SKUs. Translation will cousin a flavour name.
2. Add captions to a video with the language code that matches the new speech. BASIC 21 looks at 1×, DYNAMIC 9 at 2×. Preview the first seconds on this background.
3. Spot-check the burn against the legal string. If the spoken translation drifted, fix the audio or overlay the exact string. Do not ship a caption that says a different mg than the carton.
4. Place in safe area for the destination. A French line is often longer than the English. Two-line English can become three-line French and collide with TikTok chrome. Shorten the spoken line, or accept a smaller caption size — do not cover the pack.
5. Recaption if you then crop. Burns are pixels.

Burned-in captions are the short-form object. A sidecar SRT in English is not a French burn. Captions for silent autoplay still apply: the new market also mutes the feed.

Do not reuse the English caption file with a font change. Timing of clauses changes with the new language. ASR on the new track is the clock.

If you need authored translations on screen that nobody said — a before/after label — that is overlay copy in the target language, not ASR.

YouTube long-form AI labels sit below the player; Shorts overlays sit on the video (YouTube Blog, 27 May 2026). Labels alone do not decide monetization. Inauthentic-content rules are a different policy. A translated UGC ad that is still original remains original.

Disclose and keep the picture

Translation should not become a new generate. The point of this door is the same picture, new words.

Keep:

• The locked SKU.
• The crop you already signed.
• The duration, unless the new language is so much longer that you must extend picture as a separate job.

Change:

• The soundtrack (translate-audio-only or full dub).
• The caption pass.
• Overlay strings that are language.

If the new language will not fit, do not speed the picture to cram it. Rewrite the target script shorter, or cut a second Short. Speed-changing a talking head so Spanish fits is how visemes die even if you never asked for lipsync.

A market list is several jobs, not one. Spanish, then French, then German: each is a target_lang, a poll, a listen, a caption pass. Do not batch-burn 165 language codes onto one English file and call it localisation. The caption tool transcribes; it does not invent a market. Keep the picture parent. Swap audio. Recaption. That is the reuse silent rows were for, and it still works on a voiced master if you accept replacing the original track.

EU artistic/satirical labelling may avoid spoiling a work; a product ad is not that exemption. Provider marking grace through 2 December 2026 is a provider clock, not a reason to skip a deployer deepfake label at first exposure.

Translate the hosted file in Versely, then caption the language that is now on the track. Do not burn English and hope the market reads around it.

FAQ

How is this different from dubbing?

Same dub_video tool. translate_audio_only skips the full voice-cloned pipeline. Full dub clones delivery; heygen also retargets the mouth. Use the light path when understanding is the job.

Do I need to set the source language?

You can. Auto-detect exists. Set source_lang when the mix would fool a detector.

Can I caption in French without translating the audio?

You can burn French overlays you wrote. ASR captions of English audio will still be English (or a bad French transcript of English). If the speaker is English, French subtitles are a translation you author or a dub you run first — not add_veed_captions alone.