The dubbing chain: English, French, Spanish from one file

The Dubbing Studio examples are not three videos. They are one video: English original, then the same take through a French lip-sync, then a Spanish lip-sync, captions burned in on the translations. That chain is the product. A second shoot in French is a different company.

Dub a video is the tool. The AI dubbing tool page is the longer capability map — two engines, voice character carried over, lip-sync when there is a mouth. This post is the decision the examples are making for you: lock one master, then localise it. Do not rebuild the creative per market.

What the rail is showing

English is the source. French and Spanish are not "versions." They are the same picture, the same performance, a translated read, mouth re-timed, captions regenerated. If the French clip looks like a different person or a different kitchen, the chain failed. If it looks like the same person speaking French, the chain worked.

That is a localisation product. It is not:

• Three scripts, three prompts, three generates.
• A subtitle file on the English master, shipped to a French feed as if that were a French video.
• A new avatar "for the market" with a generic narrator.

Subtitles translate the words for people who will read them. A dub translates the watch. On short-form, most people will not read a subtitle track in a language they do not speak; they will scroll. A lip-synced dub is how the same approved take earns a second feed.

The decision rule

Answer two questions about the master. The chain follows.

Is the picture locked? If you are still regenerating the take, do not dub. Every language version derives from the master; a recut means re-dubbing everything. Dub last. Always.

Is there a mouth in frame? If yes — talking head, UGC, presenter — run the lip-synced path, which is what the example rail is. If no — voiceover over b-roll, screen recording, product turntable — skip lip-sync and dub the voice only. There are no lips to save, and you should not pay for a mouth pass on a kettle.

The rest of the rule is sequencing, not taste.

1. Approve one English (or source-language) file. Picture, read, captions, offer. The thing a stakeholder already said yes to.
2. Isolate the voice if the mix is busy. Music and SFX should be re-laid under each dubbed voice, not translated through the mix.
3. Run French and Spanish as two jobs off that file. Independent passes. A bad Spanish take does not block French.
4. Re-caption each language. Timing does not survive translation. Sentence length changes. Do not translate the English caption file and hope.
5. QC the words, not the vibe. A native reader for anything that will spend money. Organic tests can spot-check. Paid ads cannot.

Spanish and French are the example pair because they are a sane first wave for an English master, not a ceiling. Add a language when analytics already show viewers there, not when a listicle says "20 markets."

What you do not rebuild

The temptation, once the French take exists, is to "make it more French": new hook, new location, new joke. That is a reshoot. It may be the right call for a campaign that leans on wordplay or a local reference. It is the wrong default. The whole point of the chain is that the creative is already approved.

Keep:

• The crop, the wardrobe, the product placement, the caption family.
• The offer and the URL, unless the market genuinely cannot use them.
• The runtime. Padding a translation to "feel local" is how you wreck lip-sync.

Change:

• The words, including on-screen type that is part of the picture. Burned-in English on a pack shot is not dubbed by a voice pass.
• The captions, as a new burn-in, in the same family you locked on the master.
• Nothing else, until a native QC flags a cultural miss that a translation cannot fix.

If the script is a pun, the chain will give you a translated pun, which is usually a dead pun. That is not a model failure. It is a brief that was never going to localise. Write the master for the ear in more than one language, or budget a human rewrite for that market and treat it as a second master — a new chain, not a flavour of the first.

Put the languages on the same row as the master in the tracker. The moment the English file is picture-locked, the French and Spanish jobs should already be named. That is how the rail is built. A studio that "maybe translates later" is running a reshoot calendar and calling it distribution.

FAQ

Do I need lip-sync on every dubbed language?

No. Lip-sync is for when we see the mouth. Voice-only dub when we do not. The example rail is a talking-head chain, which is why the French and Spanish clips are lip-synced with captions. A voiceover explainer dubbed into Spanish should not go through that path just because the talking-head example did.

Why start with French and Spanish rather than the largest market on a map?

Because they are the pair the product actually demonstrates, and because they are a sane first wave for most English masters: real audience, similar visual grammar, enough native QC available that you will actually check the words. Start where you can finish the chain. Add languages when the file already has viewers there.

Can I dub before the English captions are burned in?

You can. You should not. Captions on the master are part of the approval. The dubbed files need their own captions anyway, in the same family. If the English burn-in is still in flux, you will redo the localisation captions twice. Lock the family on the source, then regenerate per language.

Is a dubbed clip a new video or a copy of the original?

For the platform, it is a new upload. For production, it is a derived asset: same picture, new audio, new captions. Do not prompt a new generate "in French." Do not treat the dub as a chance to recut. If the French market needs a different hook, that is a different master and a different chain.