Versely
    Guides

    Audio-only translation should wait for the master soundtrack

    dub_video with translate_audio_only is a lighter language pass. It still bills against length, still runs async, and still dies if you recut the parent.

    Versely Team••3 min read

    Audio-only translation is the lighter sibling of a full dub. Same tool: dub_video with translate_audio_only. No cloned-voice theatre, no HeyGen mouth work unless you ask for a different job. You still submit a Versely-hosted file. You still poll get_dub_status. You still inherit the engine caps — 30 minutes on ElevenLabs, 8 on HeyGen. And you still should not start until the soundtrack is the soundtrack you will ship.

    Translate a video into another language is the task. Cost scales with length like a dub. estimate_cost and check_credits before you submit. The AI dubbing tool is the surrounding surface.

    Lighter is not earlier

    People pick this path because it feels like "just the words." Words are attached to a clock. A recut changes the clock. A new take changes the line. Translation does not reposition picture. If the French VO is 12 seconds and you later trim the English master to 8, you have a translated track for a file that no longer exists.

    Full dubbing (dub a video) clones delivery and can lip-sync. This path is for when the mouth is not the product and you need the content in another language. Both paths assume a parent. Neither path is how you discover whether the English take is any good.

    If the English mix is still scratch — temp VO, wrong bed, un-isolated noise — you will translate the scratch. Then you will mix the real English, then translate again. That is the leak.

    What has to be true

    • The picture is allowed to freeze.
    • The English (or source-language) track is the mix, not a placeholder.
    • You know you do not need viseme retargeting. If you do, you wanted the full dub, still later.

    Then submit. Then wait. Then caption in the target language if the destination is silent-autoplay. Subtitling the English file and translating audio afterward is two finishing passes that fight.

    Do not use translation as a script draft

    If you needed a written translation to review, transcribe the keeper (transcribe_audio) and translate the text in a doc. An async media job is a bad notepad. The output is a new audio (or video) file, not a paragraph you can redline without watching.

    External URLs are not inputs. Ingest first. Ingesting a miss so you can translate it is extra storage plus a job plus a delete.

    FAQ

    Is this cheaper than a cloned dub?

    It is lighter-weight in work. Cost still scales with length. Do not assume "audio only" is a free preview of the Spanish market.

    Will on-screen English text get translated?

    No. This is an audio/language operation. Burned-in English type stays English. If the plate has baked type, you needed overlays you control, on a locked cut, in the language you will ship.

    Can I run this while the picture is still generating?

    You can queue anything. You should not. The job that returns is bound to the source you sent. A newer take is a different source.

    When do I pick HeyGen versus ElevenLabs here?

    Same caps as the full dub. If you do not need lip-sync, ElevenLabs on a locked master is the usual path. HeyGen's 8-minute cap is a reason to cut the parent first — which, again, is lock.