Dubbing against regenerating a language
Dubbing keeps the picture and replaces the voice, which makes the mouths wrong unless something corrects them. Regenerating produces matching mouths and a different performance, different timings, and a second set of everything to review. As of 2026-09-12.
| Route | Keeps | Costs |
|---|---|---|
| Dub the audio | The picture, exactly | Mouths that match the first language |
| Regenerate in the language | Matching mouths | A second set of everything to review |
| Subtitle only | One picture version | The audience reads |
Inclusion rule. Routes available from a finished episode. Approaches requiring reshooting live action are out of scope. Order. Alphabetical by route.
1Dubbing keeps the picture, and that is the whole trade
The shots stay identical, so continuity, framing and grade carry over untouched. Only the audio is new, which makes it the cheaper route by a wide margin and the one that keeps the language versions visually identical.
The cost is that the mouths were generated for the original language. Whether that matters depends on the shot: a wide shot forgives it, a close-up does not.
2Timing is the hidden cost of dubbing
The same line takes different amounts of time in different languages, often by a large margin. A translated line that does not fit the shot has to be shortened, which is a script decision, or the shot has to be re-cut.
Budgeting a dub as an audio task and discovering it is an edit task is the usual way this goes over. Translating for length, not only for meaning, is the fix.
3Regenerating gives matching mouths and a new review
Generating the scene again in the target language produces speech and picture that agree. It also produces a different performance, different framing decisions inside the model, and possibly a different-looking character.
That means every language version is a separate thing to check, and the continuity work multiplies by the number of languages rather than being done once.
4Which one the shot decides
Close-ups and anything where a mouth is legible push towards regeneration. Wide shots, over-the-shoulders and anything cut away from the speaker are fine dubbed. Many productions end up mixed, dubbing most of a scene and regenerating the two close-ups.
Deciding this shot by shot rather than project-wide usually costs less than either pure approach and looks better than the cheap one.
5Voice identity across languages is its own problem
Keeping a character recognisable across languages is harder than keeping them recognisable within one. Whether a voice can carry across languages at all is documented differently by different models, and sometimes not at all.
Where it cannot, the practical answer is casting a deliberate counterpart per language and keeping that pairing recorded, so a later episode uses the same one.
6Subtitles are a third route and a real one
For many markets subtitled delivery is normal and expected, and it avoids both problems entirely at a fraction of the cost. It also keeps a single picture version, which makes every later fix apply everywhere at once.
It is worth pricing alongside the other two rather than treating it as the fallback, particularly for a first release into a market whose response is unknown.
7Decide before the first episode, not after
All three routes are cheaper to plan for than to retrofit. Shot selection, line length and how much of the performance is carried by close-ups all change depending on which route the season will take.
A season shot without a language plan is effectively a season shot for dubbing, because that is the only route that does not require the picture to change.
8Names and places travel badly
Character and place names written for one language are frequently unpronounceable or comic in another. Where a season is planned for several markets, testing the proper nouns in each of them belongs at the naming stage rather than at delivery.
Where a name has to change per market, record the mapping once and use it everywhere, including in prompts, so a later episode does not reintroduce the original.
9One route per season, decided in writing
Mixed approaches work shot by shot inside a scene, but switching route between episodes does not: a season dubbed for six episodes and regenerated for the seventh will look and sound like two productions.
Write the chosen route into the production notes with the reasoning, so the decision survives a change of personnel rather than being retaken by whoever is available.
10Where the model-level facts live
Which languages a model states, which voice controls it documents and what it says about lip alignment are recorded in the table, each cell carrying its source page and check date.
This page is about which route to take. The table is about what the route requires from the model you intend to use.
11Where the published detail is logged
What is described above is how the techniques work in general. Which models document which of them, in whose words, is kept in the speech controls table with a date on every field.
- How the table is read — what counts as a documented control here
A mechanism note rather than a documented field. Nothing here is attributed to a model, and nothing here is a claim about one. The sourced material is on the speech controls table. Related: How lip-sync works, Casting a voice.