Preparing a Script So It Dubs Cleanly

By the time a video reaches translation, most of its localisation cost is already fixed. The sentences are the length they are, the figures of speech are illustrated on screen, the numbers are embedded in clauses that will not re-order, and the graphics have text burned into them.

None of that is expensive to avoid while writing, and all of it is expensive to work around afterwards. This guide covers what to change in a script that will be dubbed, why each change matters, and the specific decisions in production — gesture, on-screen text, pacing — that decide how much freedom the translator will have.

Short Sentences, One Idea Each

The single highest-impact change is sentence length, and the reason is arithmetic rather than style.

A translated sentence may be substantially longer than its source. If the sentence occupies one shot with a fixed duration, that extra length has nowhere to go — the line either overruns or has to be compressed. Two short sentences covering the same ground can absorb the difference between them, because the pause is adjustable and each half can be trimmed independently.

So the working rule is one idea per sentence. Not a word limit, which people game, but a structural test: if a sentence contains a claim and a qualification, or two things joined by "and" that could stand alone, split it.

Harder to dub:
Because the second pass depends on the first, an error introduced
early will carry through everything that follows, which is why the
review happens between stages rather than at the end.

Easier:
The second pass depends on the first. An error introduced early
carries through everything after it. That is why the review happens
between stages, not at the end.

The second version is also clearer in the original language, which is the general pattern here: writing for translation is mostly just writing well.

Numbers, Names and Units

These are the items that cause disproportionate trouble, because they are exactly where languages diverge in word order.

Numbers. Write them so the sentence can be re-ordered around them. A construction that depends on the number appearing in a particular position will fight every language whose grammar puts it elsewhere.

Dates. Formats differ, and so does whether the day or the month leads. Writing dates in a way that survives re-ordering, and avoiding constructions that hinge on the format, prevents a rewrite per market.

Units. Decide in advance whether measurements are converted per market or kept as-is. Both are legitimate; changing your mind halfway through a library is not. Conversions also change length, sometimes considerably.

Names. Supply a list with the intended pronunciation and the intended rendering per language. Left unspecified, the same product name will be pronounced three ways within one video, which is the error viewers notice most.

Watch What the Picture Commits You To

Three production decisions constrain the translator more than anything in the script itself.

Illustrated figures of speech. A metaphor accompanied by a literal image fixes that metaphor in every language. If the shot shows what the phrase says, the phrase cannot be replaced. Where localisation is likely, choose images that illustrate the point rather than the wording.

Gesture. Counting on the fingers, showing a size with the hands, pointing at an object — each of these fixes both what the sentence must say and when it must say it. The translated line has to arrive on that beat, which removes most of the flexibility that would otherwise absorb a length difference.

On-screen text. The most expensive item on the list. Text rendered into the footage has to be recreated per language, which means re-rendering the shot. Built as a separate layer in the project, it becomes a text change and a re-export.

The decision to keep text on a layer costs nothing at the time and saves the largest single expense in most localisation jobs. It is worth making even when localisation is only a possibility.

Leave Room in the Edit

The edit itself can make dubbing easier or harder, and a few habits cost nothing.

Leave a beat between sentences rather than cutting speech tight. Those small gaps are where a longer translated line finds its extra time, and an edit with none is one where every line must fit exactly.

Avoid cutting exactly on the last word of a sentence. A shot that ends a few frames after the speech gives the dub somewhere to overrun without crossing into the next scene.

Where a section is dense and fast, consider whether it needs to be. Rapid delivery in the original becomes impossible delivery in a language that needs more words, and the resulting compression is audible.

None of this is visible as a compromise in the original version. It reads as a slightly more relaxed edit, which most speech-driven video benefits from anyway.

Write the Structure, Not Just the Words

One more habit that pays across every language version: make the shape of the argument explicit in the script.

Sections with clear boundaries, signposting sentences that mark a turn, and a consistent pattern for introducing and closing each part. This helps the original viewer, and it does something specific for translation — it gives the translator a map, so they know whether a given sentence is a claim, a qualification, an aside or a summary.

Without that, translation happens sentence by sentence and the structure has to be inferred. Inferred differently in eight languages, it produces eight versions that make the same points with subtly different emphasis, which is how a series stops feeling like one thing across markets.

It also makes the QA pass tractable. A reviewer who can see where each section begins and ends can check the version against the source at a structural level, rather than comparing sentences and hoping the whole adds up — the ordering for that review is set out in a QA checklist for dubbed video before you publish.

A Preparation Pass

That last step is what makes the timing stage tractable rather than a series of compromises, and the reasoning behind it is set out in why dubbed audio drifts out of sync, and how to fix it.

The second step deserves its own attention on any script with personality in it, because figurative language is where the most difficult judgements live — covered in translating idioms and jokes for video.

What This Costs

Almost nothing, which is the point. Shorter sentences, fewer idioms, clearer numbers and text on a layer are all improvements to the original video considered on its own terms.

The one genuine trade-off is voice. A script stripped of every figure of speech can read as flat, and personality-driven content loses something real. The answer is not to remove all of it but to spend it deliberately — keep the expressions that carry the speaker's character, drop the ones that were merely habit, and know in advance which ones will need a decision per language.

For content where the personality is the product, that trade-off may point away from dubbing entirely, which is the choice examined in choosing between dubbing and subtitles per market.

Next Steps

Take the script of a video you are planning to localise and run the first step alone: split every sentence carrying two ideas. Count how many you split. That number is a fair estimate of how many timing problems you have just prevented.

Then check whether your on-screen text is on a layer or in the picture. If it is in the picture, changing that on the next project is the single largest saving available in the whole localisation process. The pipeline it feeds is described in how AI dubbing works, step by step, with the rest under dubbing and translation.