How to Localize Training Videos with Synthesia

Most “AI localization” posts skip the two facts that actually decide the project. First: you cannot localize a moving English draft. The master has to be locked — script, screens, captions style — or every language becomes its own rewrite. Second: the tool this site sends for a talking-head language cut is Synthesia (synthesia.io). This guide is the working path for L&D and enablement: freeze the English file, translate, swap the voice, decide the UI, QA, version.

When a training video needs a localized avatar — and when it doesn't

Re-render in another language when learners need to hear the lesson and you will not fly talent or book a booth for Spanish, French, and German. Onboarding, compliance refreshers, and product enablement that already exist in English are the fit. Duplicate the project, replace the script, switch the voice, generate. That is the economic case.

Skip the localized presenter if captions on the English file are enough — a US-heavy audience that wants access, not a new VO. Skip it if the market has a different policy or a different SKU; that is a new lesson, and you are back on the training how-to. Skip it if the file is a sales demo still waiting on product sign-off — localization there is a next step on the product-demo how-to. Skip it if the video will run ads on YouTube — the public avatar how-to is the other desk.

The seven-step Synthesia localization

1. 01

   Lock the English master before anyone translates a line

   Localization copies a signed lesson. Freeze the English script, the avatar, the crop, the captions style, and every screen grab. If Legal or L&D is still moving a policy number, stop. A Spanish cut of a draft is two drafts. The already-live training how-to is how you make that English file. This page starts after it exists.

2. 02

   Confirm you need a localized presenter — not captions, not a new lesson

   Re-render in Spanish, French, or German when learners need to hear the lesson in their language and you will not re-book talent. Captions on the English file are enough when the audience already works in English and you only need access. Write a new lesson if the market has a different policy, a different product SKU, or a different UI. Do not “localize” a module that is actually a rewrite.

3. 03

   Open Synthesia (synthesia.io) — check the domain

   Synthesia is the presenter desk this page is written for. Synthesys (synthesys.io) is a different company. HeyGen and Colossyan show up in the same “AI localization” search; they are editorial mentions here, not hops. Mixing the names is the expensive mistake. This page sends you to Synthesia only.

4. 04

   Pick the plan that can actually ship language versions

   Basic ($0) is an audition: 9 avatars, 10 minutes a month, no clean download. Starter ($29/month, or $18/month billed yearly) is where downloads start — that is the usual path if you will translate the script yourself and swap the voice. Creator adds minutes. Enterprise is where 1-click translation into 80+ languages and brand kits live. Unused minutes do not roll over. Video and dubbing share the same minute pool. Do not buy Enterprise for one Spanish file you can re-render by hand.

5. 05

   Translate the script. Decide whether on-screen UI stays English

   Commission a human translation of the locked English script — or a machine draft plus a native editor. Spell product names, plan names, and policy citations the way the learner should hear them. Keep a glossary so “workspace” does not become three different words. Then decide the picture: if learners use the English product, leave the UI in English and say so in the VO. If they use a localized product, recapture those screens. Do not leave English burned-in captions on a Spanish video.

6. 06

   Swap the voice and avatar language, then generate the localized cut

   Duplicate the locked English project. Do not overwrite it. Paste the translated script, switch the avatar voice to the target language, keep the same presenter and framing, and generate. Mouth-sync is consistently strong in English and major European languages — Spanish, French, and German are the usual first three. Hand gestures still loop. These are narrators. Edit the translated line and re-render instead of booking a booth.

7. 07

   QA with a native speaker, then version the files

   A native speaker watches once muted (captions and UI) and once with sound (tú/usted, tu/vous, register, false friends, numbers). Fix the script and generate again. Name the exports so English and Spanish share a version: course-id_en_v3 and course-id_es_v3. When the English master moves, bump the version and re-localize. Do not silently replace v3 English with v4 and leave Spanish on v3 in the LMS.

Localized re-render vs captions vs a booth day

“AI localization tool” is a search, not a product. The decision is the deliverable. Pricing on the Synthesia column matches the live synthesia.io/pricing page as of August 2026. The yearly toggle shows Starter at $18/month; the FAQ on the same page still quotes $264/year. We report both. Unused minutes do not roll over. Video and dubbing share the same pool.

Lock the English master

Treat the English file as source, not as a first draft you will “also translate.” Freeze the script word-for-word. Freeze the presenter, the crop, and the background. Freeze every screen grab and every card. If a policy number or a product name is still in review, the localization queue is closed. Shipping ES/FR/DE off a moving EN file is how you get three different lessons in the LMS.

If that English lesson does not exist yet, stop and make it. The working path is How to make an AI avatar training video. Come back here when L&D and Legal have signed the cut you will actually assign.

Translate the script — do not just swap the voice on English words

A language cut is a new script in the same presenter’s mouth. Machine-dump the English into Spanish and you will hear “workspace” three different ways and a formal usted in a module that should sound like a teammate. Commission a human translation, or a machine draft plus a native editor, against a glossary: product names, plan names, feature labels, and citations that must stay identical across languages.

Spell numbers and SKUs the way the learner should hear them. Keep legal citations in their original form if that is how the policy is filed. Do not ask the avatar to “sound warmer in French.” These presenters are narrators: blink cadence, micro head-tilt, a looping hand. On a head-and-shoulders framing that is enough. Full breakdown of what the avatars actually do is in our Synthesia review.

Swap the voice. Keep the same face.

Duplicate the locked English project. Paste the translation. Switch the avatar voice to the target language. Keep the same stock (or personal) presenter and the same crop so a learner who takes EN and ES modules still sees one instructor. Generate. Watch the first second, proper nouns, and policy numbers. Edit the translated line and re-render. That loop is the product. A booth re-book is what you are not buying.

Languages and voices are published as 160+ on every Synthesia tier. Mouth-sync is consistently strong in English and major European languages; Spanish, French, and German are the usual first wave for a US-written library. Budget a native-speaker pass outside that set, and budget one anyway for register and idioms inside it. 1-click translation into 80+ languages is an Enterprise accelerator after the English master is locked — not a reason to start translating a draft.

On-screen UI: keep English, or recapture

The voice is only half the picture. If the learner will click the English product, leave the UI in English and say so in the VO — “the buttons on your screen are in English.” If the market ships a localized product, recapture those screens and match the labels the learner will actually see. A Spanish track over the wrong chrome is a ticket, not a translation.

Policy and compliance modules with no product UI skip this decision. What they cannot skip: burned-in captions. Strip English captions before you export a localized cut, then burn captions in the target language. Mixed-language burnt text is how QA fails in the first ten seconds. Brand kits — locked colors, logos, fonts — sit on Enterprise. For a first ES/FR/DE wave, a stock presenter and a clean Starter download are the whole buy.

Which Synthesia plan ships the language cuts in 2026

The 2026 pricing page meters Synthesia in credits plus video minutes. Basic is $0 with 9 avatars and 10 minutes a month — downloads and logo removal start on Starter. Starter is $29/month (1,200 credits, 10 minutes of video or dubbing, 125+ avatars) or $18/month billed yearly. Creator is $89/month (3,600 credits, 30 minutes, five personal avatars, API) or $64/month yearly. Annual Starter lists 14,500 credits and 120 minutes a year; annual Creator lists 44,000 credits and 360 minutes a year. Unused minutes do not roll over. One English module plus three language cuts is four renders on the same meter.

For a first localization, start on Basic with a real translated paragraph — not a demo sentence — and watch it with a native reviewer. Go to Starter when you need a logo-off file for the LMS. Buy Creator for minutes or a personal avatar — not for “more cinema.” Buy Enterprise when you want 1-click into 80+ languages, a brand kit, SCORM, or SSO. Confirm commercial and internal-use terms on the plan you will export from. We do not invent a “full commercial rights” line the pricing page does not print.

Native-speaker QA is not optional

Mouth-sync can be fine and the lesson still wrong. A native speaker watches the export muted first: captions, cards, and whether the UI matches the market. Then with sound: formality, gendered agreement, false friends, and numbers. File line notes against the script, not against the video. You change the line and generate again.

Do not skip this on Spanish, French, or German because they are “major European.” Those are the languages where a bad tú or a calqued metaphor still ships if nobody listens. Employees still deserve a plain-language note when the presenter is synthetic — say it in the module intro or the LMS description, in that language.

Version control so the LMS does not lie

Name every export with a course id, a language code, and a version that matches the English master: onboarding-safety_en_v3, onboarding-safety_es_v3, onboarding-safety_fr_v3. When English moves, bump to v4 and re-localize the delta. Do not overwrite the English project in Synthesia — duplicate, then translate the copy. Keep the glossary next to the version, not in a chat thread.

If the 2026 plan is “one English library, five languages, next quarter,” write that as a rollout after v1 English is locked, not as the first invoice. The comparison of presenter platforms — including where Synthesys sits as a different company — lives on Synthesia vs Synthesys and Best AI Avatar Video Tools 2026. This page does not hop there. The only Try button here is Synthesia.

HeyGen, Colossyan, and a caption-only file — editorial only

Those names sit in the same “localize training video” search. HeyGen is the usual Synthesia alternative in reviewer write-ups: faster custom-avatar turnaround, punchier social presenters, less of the LMS/SSO stack. Colossyan shows up in L&D roundups as another corporate-avatar desk. A caption-only English file is the right tool when you do not need a new voice. None of them have a live hop on this page, so there is no Try button. If you need the localized talking-head we can route, it is Synthesia.