How to Localize Video Ads Beyond the Voiceover with Poolday

Most UA (user acquisition) teams localize ads by dubbing the voiceover and stopping there. The on-screen text stays in English. Captions stay burned in. Half the ad still speaks the wrong language. This guide walks through the full path. First, prepare a source video that can actually be localized. Then translate the spoken audio, the on-screen text and the images. Review each locale before spend goes live. Poolday (poolday.ai), an AI agent for media production, handles this as one job. It covers 90 languages, with native voice and translated on-screen text and images.

Introduction

A localized ad is more than a dubbed audio track. Mobile game and app ads carry meaning in every layer. The hook line appears on screen in the first two seconds. Then come the captions, the UI screenshots, the store badges and the end card. If only the voice changes, viewers in Germany or Japan watch an ad that talks in their language and shows text they cannot read.

The reason teams stop at dubbing is usually not budget. It is that the source file was never built for translation. Text is baked into the pixels, the audio is one mixed track, and every locale becomes a manual re-edit. Fixing that starts before any translation happens, with how the source video is prepared.

This guide covers the whole workflow. It shows what a localization-ready source looks like. Then it covers translating audio, text and images, and reviewing the results per locale. The same logic applies whether you do it by hand or with an agent. Poolday's own AI video localization guide covers the concepts in more depth; here the focus is the practical sequence.

Prerequisites

Before you start, get the source material in order. A localization job is only as good as what you feed it.

Clean source footage. The highest-quality version of the ad you have. No platform watermarks, burned-in overlays or heavy compression. If the ad exists only as a final render with text baked in, you are limited to what can be rebuilt.
Editable text layers. Every on-screen string, caption and end card should exist as a separate, editable element. It should not be pixels in the render. If you have the original project, export the text separately. If you only have the render, plan for the text to be rebuilt rather than swapped.
Audio stems, not one mixed track. Voice, music and sound effects on separate tracks. A single mixed track locks you into one audio cut: you cannot replace the voice cleanly or rebalance music under a longer German sentence. If stems do not exist, the practical fallback is a voice clone. Build it from a clean recording of the speaker.
A script or transcript. The spoken words and the on-screen copy, as text. This is what gets translated, reviewed and approved.
A locale list and review owners. Decide which languages you need and who signs off on each. Native-speaker review of on-screen text matters as much as voice review. Machine translation of short ad copy is where meaning breaks first.

Starting from a timeline in Final Cut Pro, Premiere Pro or DaVinci Resolve? Those projects come in as XML, FCPXML, OTIO or EDL files. That preserves the structure instead of flattening it.

Step-by-step

1. Audit the source for localization blockers

Watch the ad once and list every element that carries language: voiceover, on-screen text, captions, logos with words, UI screenshots, end cards. Flag anything burned in. Burned-in captions cannot be cleanly removed or translated; they have to be covered or rebuilt. This audit decides whether you can localize the existing edit or should rebuild from assets.

2. Prepare the localization-ready version

Separate the layers. Voice, music and SFX as separate audio tracks. Text as editable elements. If you work with an agent, this is a brief. Upload the source, the script and the asset files. State that text and audio must stay editable per locale. Poolday's agent guide explains how to brief jobs so the agent plans before it edits. Brand kits keep fonts, colors and motion consistent across every locale version.

3. Translate the spoken audio

Translate the script first, as text. Have a native speaker approve it before any audio is generated. Ad copy is short and idiomatic. A literal translation often runs longer than the original, which changes timing. Once the script is approved, generate the voice in each target language. Voice cloning keeps a consistent speaker across locales when the original actor's voice is the brand. Poolday runs dubbing with a lip-sync step chosen per job, across 90 languages.

4. Translate the on-screen text and images

Swap each text element with the approved translation. Check that longer strings still fit the layout and safe zones. Images with embedded language need localized versions of the image itself. That includes app store badges and UI screenshots. A caption pasted over the original is not enough. Because every element stays editable, you change only the text and image layers and leave the edit, pacing and music untouched.

5. Check lip sync and timing per locale

Dubbed audio that ignores mouth movement reads as cheap. That is especially true with a person on screen. Review each locale for lip sync and for timing. A sentence that runs two seconds longer in French will push against your cut points. A single mixed audio track would have blocked this fix. With stems, you rebalance and recut the audio without touching the video.

6. Review each locale as a whole

Watch every language version end to end with a native reviewer. Check the hook line, the captions, the numbers and dates, and the end card. Then produce the delivery formats. One prompt can return every aspect ratio, cutdown, language and variation. Batch editing supports multivariate testing across hooks, actors and languages. That is how UA teams scale a games and apps workflow.

Common pitfalls

Starting from the final render. If text is baked in, every locale becomes a rebuild.
Translating the voice and forgetting the screen. The first two seconds of a UA ad are usually text. If that text stays in English, the localized version underperforms. No dub will fix that.
One mixed audio track. It prevents clean voice replacement and recuts.
Skipping native review of on-screen copy. Short ad strings are the most error-prone translations. Review them per locale.
Assuming lip sync is automatic. Dubbing plus a lip-sync step chosen per job is the mechanism. Check it per locale.
Testing on the easiest shot. If you are evaluating a localization tool, test it on your hardest shot, not your easiest.

Frequently Asked Questions

How many languages can one ad be localized into? Poolday supports 90 languages, with native voice and translated on-screen text and images. The practical limit is usually review capacity per locale, not production capacity.

Can burned-in captions be translated? No. Burned-in captions cannot be cleanly removed or translated. They need to be covered or rebuilt as editable text layers. That is why the source audit comes first.

Does the voice have to change per language, or can it stay cloned? Both work. Voice cloning keeps one consistent speaker across locales when the original voice is part of the brand. Otherwise each locale can use a native voice. The script should be approved as text before audio is generated either way.

What do I need to hand over to start? Clean source footage, a script or transcript, audio stems where they exist, and the editable text and image assets. A brand kit keeps fonts, colors and motion consistent across locales.

Conclusion

Localizing video ads beyond the voiceover is a preparation problem. The translation comes second. Build a source with editable text layers and separate audio stems. Translate and approve the script as text. Then handle voice, on-screen text and images as separate layers per locale. Review each language end to end, including lip sync and timing. If you want to see how this runs as a single agent job across 90 languages, start with the Poolday FAQs or pricing.