How to Use HeyGen to Translate a Product Video

The short answer: build a brand glossary for each target language, upload the product video, set the source language and up to 10 targets, pick the engine by how the speaker faces the camera, translate one language first, proofread the script on Pro or above, let an agent watch the queue while a native speaker reviews every result, and share each version with captions.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

A demo recorded once in English and re-voiced in the presenter's own voice, lips matched, replaces a reshoot per market. That is the use case where AI video pays for itself fastest. HeyGen documents the flow in detail, including where it breaks: a speaker turned away from the lens, two people talking over each other, text burned into the picture. For plans and credits, see the HeyGen glossary entry.

Check the source before you spend a credit:

• One spoken language. A video that mixes languages fails or comes back with an inaccurate transcript, per HeyGen's translation guide.
• Faces toward the camera. The same guide asks for speakers within 45 degrees of the lens, one at a time, in close-up, with little background noise.
• A plan that fits the length. Free stops at 1 minute; Creator, $29 a month with 600 credits, and Pro cap each video at 30 minutes; script proofreading starts on Pro (pricing page, checked on 24 October 2026).
• A native speaker per market. Somebody who will watch the whole result before it ships.

1. Build a glossary for each language

Click Brand in the left navigation, open the Brand Glossary tab and choose Create Brand Glossary, then upload a CSV or add terms by hand, as the brand glossary article describes. On the Translations tab, mark product and company names Don't Translate, and use Force Translate where a term must always come out one way. HeyGen recommends a separate glossary per language, because the rules differ.

2. Import the video

Click Translate in the left menu. Drop in an MP4, MOV or WEBM file of up to 5 GB, pick an existing project, or paste a YouTube or Google Drive link; Drive files must be shared with anyone who has the link, and Vimeo links are not supported (translation guide).

3. Set source and targets

Leave the source on Auto Detect or set it by hand, then choose up to 10 target languages for the job, per the translation guide. Many languages come in two forms. The base form keeps the speaker's own accent, which the guide suggests for interviews; the localised variant, English American for instance, gives a native accent and is the one it suggests for demos, tutorials and onboarding.

4. Pick the engine by the footage

Audio Only costs 4 credits a minute and moves no lips. Speed costs 6 and lip-syncs front-facing speakers; Precision costs 10 and copes with side profiles, hands or microphones in front of the face, camera changes and several speakers, per the translation guide. For example, an 8-minute demo into three languages costs 8 x 3 x 6 = 144 credits on Speed and 240 on Precision.

5. Open Advanced

Select the glossary for this language. Dynamic Duration, on by default, lets segments stretch or shrink by up to 20% so the new language sounds natural, which can change the video's length (translation guide). Remove Background Sound strips the music; sound effects are always removed from the translated audio. Enable Captions gives you two files, burned-in and clean, with subtitles as SRT only. Enhance Voice lifts clarity but can shift the tone.

6. Translate one language, then the rest

The translation guide advises running one language first and checking it before the batch, so you do not spend credits on versions that need rework. Click Translate; an email arrives when it is done, so the page can close. A source language HeyGen does not support fails at once and costs nothing.

7. Proofread before the final render

On Pro, Business and Enterprise, choose Review & Edit instead of Translate, per the proofreading article. In the editor you can reassign a line to the right speaker, change a voice, rewrite a segment, insert a line with Add a Script, and respell a word under Pronunciation. The Preview tab plays the new audio over the original picture without lip sync, so you can check the script before paying for the Result. Mark a narrator who never appears on screen as Off-Screen, or their voice gets synced to someone's face. Every save becomes a version you can restore.

8. Let an agent watch the queue

HeyGen runs a remote MCP server: add a custom connector with the URL https://mcp.heygen.com/mcp/v1/, sign in with OAuth, and its tools draw on your plan's credits. The translation tools create jobs with one ID per language and report each job's status, output link and failure reason. Ask the assistant: list today's translation jobs, say which failed and why, and draft a two-line note to each market lead with the link to review.

The human check: the agent reports and drafts, and the native speaker watches the complete video and signs it off before any link leaves the team. HeyGen's developer index sizes the OAuth route for small, one-off runs and points batches and pipelines to an API key, and its credits article says plan credits do not pay for Zapier, Make or n8n runs; those need API credits.

9. Share it with captions

Click the finished video to open its share page, where you can edit and publish captions, collect comments, which cannot be switched off, and download the version with or without captions (sharing article). The page's insights report views, average watch time and completion rate.

Where HeyGen says translations go wrong

From the translation guide and the proofreading article:

• Text in the picture. Only speech is translated; slides, graphics and baked-in captions stay as they were.
• Singing and music. Results are inconsistent, so avoid sources where they carry the content.
• A head turned away. Past about 30 degrees, or with the speaker small in the frame, lip sync slips.
• Long webinars. Over your plan's per-video cap, trim the video or split it into parts.
• A moderation stop. Email the video ID to the address in the translation guide for a manual review.

How to know it worked

Take two baselines first: what the last localised version cost, in agency fees or a reshoot, and the completion rate of the original on the share page. Then compare each translated version's completion rate against the original, and count the corrections your native reviewer makes per video, which should fall as the glossary fills.

Check the engine choice on that first language too. A menu whose top tier is called Precision makes the dearer option sound like the careful one, and it costs two thirds more than Speed per minute. If the demo is front-facing, Speed was built for it.

FAQ

How many languages can one job hold?

Up to 10, per the translation guide. Several at once land in a subfolder of Projects.

Does a failed translation still use credits?

An unsupported source language fails instantly and is not charged. For an inaccurate result, the translation guide asks you to contact support, which can recharge some credits.

Can I translate a 45-minute webinar on Pro?

Not in one piece. Pro caps each video at 30 minutes whatever your credit balance, so split it into two parts, or use Business, which allows 60, per the translation guide.