An explainer video works because three things land at the same moment: the narration, the motion, and the on-screen text. A line of voiceover arrives precisely as an icon scales up. Scene four runs nine seconds because the English script takes eight and a half. Every duration, every text plate, every pause was cut to fit one language.

That is why explainer video localization is rarely a translation problem. Running video translation on the script is the easy part. What breaks is the fit: a text plate sized for "Get started in minutes," a narrator paced to hit a beat at 00:06, a three-step diagram whose labels sit on one line because in English they fit on one line.

This guide covers that fit: designing the source explainer for localization, text expansion in German, Finnish, and Russian, when to re-render motion graphics instead of overlaying text, voiceover pacing against animation beats, cultural adaptation, source file hygiene, the sync review, and how to run a library of explainers rather than one video.

Why explainer video localization breaks animation timing

The animation is cut to the script

Animators do not animate meaning; they animate the words they were handed. When a script reads "Connect your tools in one click," the animator builds a reveal around 29 characters and times it to the read. Swap in a German translation and the read gets longer while the reveal does not.

What fails first

The first failures are invisible in the delivered files and obvious in the finished video. Narration runs four hundred milliseconds past a cut, so the voice is still talking over the next scene's music sting. A text plate wraps to two lines and the second is clipped by a mask drawn tightly around one. A button has been re-lettered but not grown, so the padding is gone.

Four failure classes

  • Timing overflow: narration or a text reveal exceeds its scene and pushes content across a cut.
  • Layout overflow: text clips, wraps badly, or escapes a container sized for the source string.
  • Semantic drift: on-screen text and narration stop agreeing because one was updated and the other was not.
  • Cultural mismatch: imagery, units, or color conventions read as noise in the target market.

The cheapest fixes are applied before the first render.

Designing an explainer for localization before the first render

Text space budgets

Size every text container to hold roughly 35 percent more characters than the source string, and check the layout at that width, not at the width of the English. Widen containers instead of shrinking type: a plate that looks generous in English looks correct in German, while a plate that fits English exactly forces a font reduction that flattens the hierarchy you built. Leave a line of vertical slack under any text block so a string that wraps to an extra line does not collide with what sits below.

Timing headroom

Put slack in scenes, not in the total runtime. Letting each scene run 10 to 15 percent longer than its narration needs is usually invisible; adding the same slack to the end of the video is not. Where a cut is locked to music, move the slack to the preceding scene. Narration that ends flush against a cut has nowhere for a language with different word order to put its final verb.

Layered project structure

Keep type on separate layers from the artwork beneath it, and keep numbers, currency values, and chart data editable rather than drawn. A graphic containing a word, a sign, or a device screen should be its own composition so it can be replaced wholesale. Use realistic placeholder text rather than dummy Latin, because your layout decisions depend on character count.

Text expansion in German, Finnish, and Russian

Planning ranges, and why they mislead

As planning assumptions, German runs roughly 20 to 35 percent longer than English in character count, Finnish 30 to 40 percent, and Russian 10 to 20 percent, with Russian's expansion concentrated in compound noun phrases that can far exceed the average. Treat all three as directional. The number that matters comes from your script, because expansion depends on sentence structure rather than on the language. A short imperative may gain one word; a compound-heavy noun phrase may double.

Where expansion actually hurts

Character count is not the binding constraint. Here is the order in which things break:

  • The width of a single line inside a fixed container.
  • The syllable count a narrator must fit into a scene.
  • The number of lines a block may occupy before it collides with something.
  • The number of text reveals the animation performs.

Syllables surprise teams most. German and Finnish pack more syllables into the same word count, and voiceover timing tracks syllables more closely than words per minute, so a four-word German line can read noticeably longer than the English it replaces.

Tactics when a line will not fit

Work down this list before committing to a re-render:

  1. Rewrite the translation into a shorter construction with the same meaning.
  2. Break the line at a different point and rebalance it so neither line becomes an orphan.
  3. Reduce tracking by two or three percent before reducing size.
  4. Widen the container if the composition allows.
  5. Extend the scene if the cut is not musically locked.
  6. Re-render the scene with adjusted animation.

Re-rendering motion graphics versus overlaying localized text

When a re-render is the only correct option

Re-render whenever the text is transformed. Rotated in 3D, mapped onto a curved surface, part of a camera move with parallax, sitting under a glow or gradient that crosses it, or revealed by an animated mask, the text cannot be replaced by a clean overlay, which would sit at the wrong angle and ignore the perspective. Text baked into a device screen belongs here even when it looks flat, because it moves with the object.

When an overlay is defensible

An overlay works when the text plane is axis-aligned, static for the entire shot, free of effects that cross the type, and unobstructed by moving elements. Under those conditions you can composite localized text in the edit, keep the source render, and skip the re-render. The conditions are narrow, and they should be verified scene by scene rather than assumed for the whole video.

The hybrid pattern

Two or three hero scenes carry camera moves and 3D type; a dozen flat scenes carry labels, callouts, and numbers. Re-render the hero scenes properly, composite the flat ones, and pair the result with a localized caption track. Subtitle translation is mechanical once the scene list exists, and a reviewed caption track also covers the viewer who watches with sound off.

Voiceover pacing against animation beats

Two clocks on one timeline

An explainer runs on two clocks. The animation timeline is fixed and quantized to frames and cuts. Speech is not: a narrator reads a line in whatever time it takes, and a good one will stretch or compress it to land where you ask. Mark anchor points rather than matching every frame. Anchors are cuts, on-screen text reveals, the logo, and music hits. A narrator given three anchors in a nine-second scene sounds natural and still lands; a narrator asked to hit every beat sounds mechanical.

Adjusting the voice without re-timing animation

  • Record to picture with anchors marked in the script.
  • Translate to a syllable budget: give the translator the scene duration and the number of syllables that fit, and let them rewrite.
  • Nudge the speech rate slightly. Small adjustments pass unnoticed; larger ones make the read sound hurried.
  • Trim silence between sentences, never inside them.

Adjusting animation without wrecking the voice

Sometimes changing the picture is cheaper than changing the voice. Extending the hold on a scene's final frame by half a second is inexpensive in most rigs. Re-timing a rigged character animation is not: it changes the easing, the secondary motion, and anything physics-driven, and it looks wrong unless an animator does it deliberately.

Human talent, synthetic speech, and cloned voices

For a hero explainer, a native actor recording to picture remains the most reliable option, which a proper video dubbing workflow supports. Where the budget does not allow a session per language, or where the asset is internal training rather than a campaign, speech generation produces a serviceable track, and a cloned voice keeps the same speaker across every language when that speaker has authorized the use. The trade-off is prosody: synthetic reads need spelled-out numbers, expanded abbreviations, and phonetic spellings of brand names, and they will not improvise around an awkward beat.

Cultural adaptation: charts, currency, icons, and right-to-left layouts

Numbers, units, and charts

Decimal and thousands separators invert between locales, so a chart template that hardcodes US convention is wrong in most European markets. Date formats, week start days, and currency placement change as well. Convert units for distance, temperature, weight, and screen size, and re-check any figure rounded for readability in the source. Translate axis labels first, since they are most likely to expand out of the plot area, then confirm that bar labels still fit. In Chinese financial contexts, red signals a gain and green a loss, the reverse of US convention.

Icons, gestures, and imagery

  • Hand gestures carry different meanings across markets, including the thumbs-up, the OK sign, and any beckoning motion.
  • Script-shaped icons do not travel: a checkmark reads differently where a circle denotes correct.
  • Objects are regional artifacts: mailboxes, wall outlets, keyboards, and phone silhouettes place or date a video.
  • Human representation needs a decision rather than a default, from skin tone to dress to body type.
  • Symbols with legal or political weight, including flags, maps with disputed borders, and religious imagery, are worth flagging to legal before the first render.

Right-to-left: mirror layout, not content

For Arabic and Hebrew, mirror the layout: reading order, progress indicators, sequence timelines, "next" arrows, and two-column comparisons. Do not mirror anything with a real orientation, including maps, clocks, photographs, product shots, logos, and interface screenshots; playback controls follow platform convention. Numerals stay left-to-right even inside right-to-left text, which produces bidirectional runs that belong to the text layer, not the layout. Mirror a walk cycle that indicates forward progress, but not a mechanical action with a real direction.

Source file hygiene and version control

Fonts and licensing

Confirm before the first render that the license covers embedded use in localized derivatives and that the family covers the scripts you plan to ship. Most display faces used in motion graphics are Latin only. If the face cannot set Cyrillic, Greek, or Arabic, decide now whether localized versions substitute a different face, which changes the identity of every title card, or whether the original should be set in a broader family. Keep the font files and license documentation with the project rather than a contractor's hard drive.

Editable text layers

Do not convert type to outlines, rasterize text, or bake lettering into illustrations. Every string that will be translated should be live text on its own layer, using a defined style, aligned to a container that can resize. The most localization-hostile technique in motion design is the character-count-driven reveal, where words appear one at a time as a counter advances. Mask reveals and position-based entrances survive translation; word-by-word reveals require a re-render every time.

Version control and handoff

Freeze the source-language master before localization starts, then branch per language. Name files predictably, in the form project_language_scene_layer_vNN, and keep a manifest recording the translation revision, the voice session date, the render date, and the reviewer. A complete handoff includes the project, fonts, source script, approved translation with a timing column, audio stems, and a reference export. Teams running this at volume should read the documentation for how subtitle and audio jobs are structured, which matters when you are generating tracks for twenty languages rather than two.

The review pass: checking sync between voice, animation, and on-screen text

The three-track check

Review with the waveform, picture, and on-screen text visible at once, and watch the whole piece at speed before stopping to fix anything. Stopping early hides failures that only appear in flow, such as a narrator consistently a beat behind or a scene that feels rushed.

A scene-by-scene sequence

  1. Confirm narration finishes at least a quarter second before each cut.
  2. Match every on-screen number against the number spoken.
  3. Check that each text reveal lands within about a third of a second of the corresponding phrase.
  4. Look for orphan words, awkward hyphenation, and lines that end on an article.
  5. Verify the call to action is legible, correctly lettered, and points to a localized destination.
  6. Confirm captions match the dub word for word.
  7. Check every element against the title-safe area, which some languages need wider.

Loudness and captions

Normalize every language to the same loudness target so switching languages does not change playback level. Broadcast delivery typically targets -23 LUFS under EBU R128 or -24 LKFS under ATSC A/85; web delivery usually sits higher. Measure the integrated loudness of each finished mix rather than trusting the session. Generate captions from the final localized audio, not the source script, so the text matches what is said.

A repeatable explainer video localization workflow

Build a scene kit, not a video

If you produce more than a handful of explainers, stop treating each as a one-off. A scene kit contains the recurring building blocks: logo bumper, problem statement, three-step mechanism, product UI montage, proof, call to action. Localize the kit once per language, including fonts, text styles, chart templates, button shapes, and transitions. Assembling then replaces recreating, and terminology stays consistent.

Batch by language

Group work by language rather than by video. One voice session can cover ten explainers when the scripts are prepared together and share a pronunciation guide. One render pass per language amortizes setup. One reviewer per language preserves terminology decisions instead of re-litigating them asset by asset.

Tiered investment

Not every asset deserves a full re-render and a fresh recording. A three-tier model keeps spending honest:

  • Tier one: hero explainers. Re-render animated text, record native voice, full QC.
  • Tier two: supporting clips. Re-render only the plates that carry meaning, reuse a synthetic track, lighter QC.
  • Tier three: archive and internal assets. Leave the picture alone, publish localized captions and a translated transcript, and accept that on-screen text stays in the source language.

Assign the tier once, record the decision, and work out the per-minute economics before committing. The pricing page lays out how finished minutes and language volume are counted.

Frequently asked questions

How much extra scene time should you budget for German or Finnish?

Start with 10 to 15 percent per scene, and add more where the script is noun-heavy or the on-screen text is long. Measure against a real translation, because a short sample understates what a full script does.

Can you skip the voiceover and localize subtitles only?

Sometimes, and for archival or internal assets it is often the right call. The catch is that a narration-led explainer loses its argument without the voice, because the on-screen text was written as emphasis rather than as the complete message. Text-led animation survives subtitles; narration-led animation usually does not.

Do you have to re-render every motion graphic for every language?

No. Re-render only where the text is transformed by a camera move, 3D, a glow, a gradient, or an animated mask. Flat, static, unobstructed text can be composited over the source render, which is where most of the savings come from.

What should you ask the original studio for?

The project file with live text layers, the font files and their licenses, the source script, a textless render of every scene with the on-screen text removed, the audio stems, and a reference export. The textless render is the most valuable deliverable, because it makes overlaying localized text possible without a re-render.

How do you handle right-to-left languages?

Mirror the layout and the directional motion, not the content. Maps, clocks, logos, product shots, interface screenshots, and playback controls stay as they are, and numerals stay left-to-right inside right-to-left text.

Is a cloned voice appropriate for explainer narration?

It can be, provided the speaker has authorized the use and the markets you ship to accept synthetic narration. The benefit is consistency: every language version sounds like the same brand voice, which matters more in a library of explainers than in a single campaign film.

Conclusion

The decision for most teams is not whether to localize but how much of the animation to rebuild. That decision is cheapest to make in the source file, before the first render, and most expensive after a language version has already shipped with clipped text and a narrator talking over a cut.

If you produce more than a few explainers a year, the order of operations matters more than any individual technique. Fix the source file first, build a reusable scene kit, batch work by language, and reserve full re-renders for the assets that carry the brand. Everything else can run on composited text, generated speech, and reviewed captions.

Start with one video. Take its project file, strip the text layers, generate a reference translation with a timing column, and see how far the existing animation stretches before it breaks. That test tells you what your library actually needs, and you can talk to the team about the pipeline once you know the answer.