AI Video Dubbing: Translate Talking-Head Video Without Reshooting

Captions are not a dub

Captions are honest and cheap. They also ask the viewer to read. On silent autoplay — which is most feeds — a mouth still saying English while the caption is German looks like a mistake. If the face is the product, budget for a dub and a lip sync pass. If the video is mostly a screen recording, captions may be enough.

AI video dubbing on LipSyncing AI is automatic dialogue replacement: translate the speech, synthesize a new vocal, rebuild visemes so the speaker still looks like they said it. It is not a prestige-series booth. It is the fastest honest way to localize a talking head, a product walkthrough, or a course chapter. Synthesia-class teams care about this split. So should you.

Dubbing vs translator vs video lip sync

Open AI Video Dubbing when you think in post terms: picture is locked, language must change, lips must follow. Open AI Video Translator when you think in tools: a language pair and a generate button. Under the hood both fold transcription, translation, voice, and a mouth pass. The pages exist because the searches exist.

Open Video Lip Sync when the language is already correct and you only need a new take or a cleaned vocal on the same picture. Open Multilingual Lip Sync / Lip Sync for TikTok or YouTube when the job is a new avatar line in another language, not an existing tape. Do not dub a pet meme through the people dubber — translate the joke, then run Talking Dog or Talking Cat.

How to dub a talking head without a reshoot

Picture lock first. If you are still cutting b-roll, wait. Dubbing a moving edit wastes credits every time the cut changes.

Do not dub the whole timeline of a montage. Pull the talking-head section, localize that, and intercut the original b-roll in your editor. You will spend fewer credits and the mouths will only run on frames that have mouths.

• Upload a talking-head clip with clean dialogue and a visible mouth. Mix the vocal louder than the bed.
• Set source and target language. Do not trust auto-detect on mixed-language tapes — set it.
• Review the translated script. Fix product names, legal terms, and jokes before you render.
• Keep lip sync on when the face fills the frame. Turn it off when the mouth is a postage stamp on a slide deck.
• Generate. Proof the first claim and the CTA. That is where a bad translation costs money.
• Export the MP4. Burn captions if the market wants both a dub and a safety net.

Quality checklist for a dub you can ship

A good dub is boring in the right way: names survive, lips follow the new language, and nothing in the mix fights the new vocal.

Have one native speaker watch the first twenty seconds. They will catch a name, a joke, or a CTA the model smoothed over. That pass is cheaper than a refund conversation with a market lead.

• Proper nouns, SKU names, and URLs were checked in the script step.
• Spanish (or Hindi, or Japanese) mouth timing does not sit on leftover English visemes.
• Voice gender and energy roughly match the on-screen person unless you chose a narrator on purpose.
• No double dialogue: original language is gone, not ghosted under the dub.
• Cuts still make sense. A language that runs longer may need a slightly longer tail — that is extra seconds on the meter.
• Legal, medical, and financial lines got a human pass. The model is a draft there.

Credits for dubbed durations

Dubbing is billed like every other generate: one credit = $0.01 on a plan; first 30 seconds = 150 credits ($1.50); extra second = 5 credits ($0.05). A 45-second localized cut is 225 credits. A 60-second cut is 300 credits. A 90-second chapter is 450 credits (150 + 60×5).

Target language does not change the second price. Resolution does not. If the translation runs longer than the source, you pay the output duration, so keep the translated script tight. A 28-second English explainer that becomes 34 seconds in German leaves the 150-credit block and starts adding 5 credits a second.

No free plan. Local preview still works. Packs are $0.08 per credit and never expire — useful when you localize twelve SKUs in a week. Yearly is 30% off, no auto-renew. Subscription credits spend first, then packs. Unused plan credits roll over while the plan is active and cancel if you cancel.

Rights on translated faces and voices

You still need rights to the person on camera and to the source audio. Translation does not launder a clip you were not allowed to use. A cloned voice in the target language needs permission for that clone, not just for the original booth session.

If the on-screen person only cleared an English internal video, stop. Localization is a new use. Get the extra line in the release or generate an original avatar in the target language instead.

• Own or license the source video, including talent releases that cover localized versions.
• Do not dub a competitor's ad and republish it with your URL.
• Music beds in the source may be copyrighted separately from the dialogue.
• Employee talking heads used in another country may need extra HR/legal review.
• Mark AI-altered media where the platform requires it.

Common dubbing failures

Most misses are source problems or skipped script review. The engine cannot lip-sync a sandwich review or invent a trademark you spelled wrong.

A clean talking-head lockoff plus a proofed script beats a “smarter” model. If you keep missing, your tape is the problem: food, profiles, crosstalk. Recut the source, then dub.

• Food, cups, and hands on the mouth: those frames have nothing to viseme.
• Overlapping speakers and podcast crosstalk: split tracks first.
• Extreme profiles and whip-pans: lock a talking-head take instead.
• Unreviewed names: the translator guessed; you shipped it.
• Captions-only jobs billed as dubs: you paid for a mouth pass you then ignored, or you skipped the mouth pass and the feed looks broken.
• Dubbing a montage: cut to the talking-head section, localize that, intercut b-roll in the editor.