HeyGen video translation for product demos

Quick answer: HeyGen Video Translation can translate a demo’s spoken narration and, depending on the mode, work with the visible speaker’s lip movement and captions. It does not translate product-interface text, graphics, or captions already baked into the source video. Treat those visual elements as a separate localization job before you publish a translated demo. Source: HeyGen’s Video Translation guide.

Important constraint: A translated voice saying “Open Settings” can be paired with a screen that still shows “Settings” in the source language. Review the real localized interface, the spoken script, captions, and timing together; an audio translation alone cannot make that scene accurate for its new audience.

Consider a software team with one 8–12 minute narrated screen recording. The video has one speaker, a few on-screen product names, cursor actions, and several visible menu labels. The team must decide which parts can be translated by Video Translation and which need a new screen capture, an edited graphic, or a human language review. The workflow follows HeyGen’s documentation checked on October 3, 2026.

Inventory the scene before choosing a mode

Watch the source video once with the sound on and once with it off. The first pass identifies narration and pronunciation-sensitive names. The silent pass reveals information a voice translation cannot change: interface labels, alert messages, hard-coded captions, and text inside graphics. Note every moment where a spoken instruction points to a specific visual element. That is where a mismatch would be most confusing.

A sentence spoken in another language is only part of a localized demo. If the screen shows a different menu name from the one the narrator says, a viewer may look for a control that does not exist in their product version. Use the actual localized application when possible. When that interface is unavailable, label a conceptual illustration honestly instead of passing it off as a capture of the shipped product.

For a long walkthrough, make the inventory time-coded: for example, 02:14–02:28 — narrator names Export; menu label visible; cursor clicks second item. This makes review finite. A language reviewer can compare the translated line, the localized label, and the click at a known timestamp rather than searching the entire video for vague inconsistencies.

Pick Audio Only, Speed, or Precision for the footage you have

HeyGen’s Video Translation guide says Audio Only disables lip sync. Speed and Precision enable it. HeyGen describes Speed for simpler frontal speaker footage and Precision for harder visual cases such as occluded or profile faces and complex conversations. Those are vendor recommendations about workflow fit, not evidence that any mode will look natural on your recording.

The middle and last rows still require a real output review. A mode name cannot establish pronunciation, timing, or visual quality. If the demo alternates between presenter and screen capture, the decision may depend on the sections where the face is visible. A screen-only segment still has its UI-localization work even when the presenter segments use lip sync.

Check the source and target languages against HeyGen’s current language list and the choices available for your selected engine before submitting. The published list can change, and account or workflow availability should be confirmed in the live tool.

Use terminology and script review deliberately

Product demos often contain words that general translation does not resolve by context alone: a feature name, a tier label, a button that keeps its English product name, or a technical phrase with an approved local equivalent. HeyGen documents Brand Glossary and Proofread as controls for terminology and script review in its translation workflow. Build a short terminology list from the scene inventory before submission. It should distinguish names that remain unchanged from UI terms that must match the localized application.

Proofread matters when the exact spoken instruction is consequential. HeyGen lists it for Pro, Business, Team, and Enterprise plans and describes word-level review. Check your current plan entitlement before assuming the control is present. If the source says “Select Export CSV” and the target UI calls the action by another approved term, a reviewer should resolve that mismatch in the target script, not simply accept a plausible translation.

An input SRT can help supply timing or text context, but HeyGen says it is a reference and may be translated again. It is therefore not a guarantee that a preapproved translated line will be spoken verbatim. In the Proofread workflow, HeyGen documents an output SRT script that can be spoken as written. If exact target wording is a requirement, verify that you have that workflow, review the line there, and compare it with the generated audio afterward. Source: HeyGen Video Translation guide.

For the 8–12 minute example, divide review into short scenes that each cover one action or explanation. The reviewer can confirm the target language’s product term, the visible label, and the click before moving to the next scene. That keeps a long video from becoming a single opaque “approve translation” decision. It also creates a useful change list if the team has to recapture only two interface screens rather than the entire walkthrough.

Treat captions as a separate deliverable

There are three different caption cases in this workflow. Hard-coded source captions are pixels in the uploaded video, so HeyGen does not translate them as text. Generated translated captions are an optional output control. A separate SRT file is useful when the final player or editing workflow needs captions it can turn on and off. Confusing those cases can leave two languages on screen at once.

HeyGen’s Video Translation guide describes captioned and non-captioned output versions and .srt export in relevant workflows. Its caption guide says projects without Proofread must select captions before submission; a Review & Edit path can update caption text with script changes. A translated video’s share page may offer toggleable closed captions. Controls and export options depend on the workflow, so check the output options in the selected plan rather than assuming every caption style and file is available in every mode.

Prepare a clean source without burned-in captions if you can. If that is impossible, plan an edit that removes or covers the old captions before adding translated ones. Then review caption timing against both the new speech and the visible screen action. A line that is linguistically correct can still be unhelpful if it appears after the cursor has already moved to a different control.

Allow for timing and source-file constraints

HeyGen documents Dynamic Duration, which can change segment duration by up to 20% in either direction. This is a segment-level planning consideration, not a promise that an entire ten-minute video will always stretch or shrink by precisely that percentage. Where the narration refers to a cursor click or a changing panel, review the translated scene against the visual timeline. If the timing no longer lines up, adjust the source edit or the localized version rather than assuming the spoken translation will fit the original cut exactly. Source: HeyGen Video Translation guide.

The same guide lists MP4, MOV, and WebM inputs up to 5GB, with an input-duration range from two seconds to five hours. The applicable per-video plan cap is narrower for many accounts: 30 minutes on Creator and Pro, 60 minutes on Business, and five hours on Enterprise. The proposed 8–12 minute demo is below those listed caps, but a longer file may satisfy the general input rule while exceeding its plan’s limit. Extra credits do not remove a per-video cap. Check the current limit for your account and chosen mode. Keep an editable master outside the translation upload so you can replace stale UI footage or captions without trying to repair text that is already baked into a finished video. Source: HeyGen Video Translation guide.

HeyGen’s published translation rates differ by engine. For one target language and one pass of the proposed demo, the arithmetic is:

The selected engine, current in-product estimate, and any other enabled features determine the useful budget for your job. These are translation credit units, distinct from Avatar III or IV generation rates; compare the live plan offer separately if you need a dollar budget. Source: HeyGen Video Translation guide.

A review path for one localized demo

1. Prepare the source. Inventory visible text, cursor-linked narration, hard-coded captions, and rights-cleared audio. Keep an editable original and, where possible, remove captions that would otherwise remain in the source language.
2. Confirm support and limits. Check both languages on the current support page, the account’s per-video duration allowance, accepted file type, file size, and displayed credits for the selected engine.
3. Choose the translation workflow. Use Audio Only when the scene does not need lip movement; examine Speed or Precision for a visible speaker under the source conditions HeyGen describes. Decide whether Proofread and a Brand Glossary are needed for controlled product terms.
4. Localize the visible product. Recapture real localized screens or edit approved graphics. Match every spoken action to the visible label and cursor step in the target version.
5. Review speech and captions. Have a target-language reviewer check meaning, product terms, pronunciation, subtitle text, and timing. Compare the delivered SRT or captioned version with the final video, not just the draft script.
6. Approve each language version separately. Recheck scenes changed by timing adjustments, UI replacements, or script edits. Keep the approved source, terminology list, and final outputs together for later product updates.

HeyGen can handle the speech-centered part of this workflow according to its documentation. The team still owns the truth of the demonstration: the interface shown, the product language used, and the synchronization between words and actions. For related decisions about creative AI workflows, the AI tools hub collects guides by the job they need to do. This page’s recommendation remains conditional on your source footage, language support, plan controls, and a review of the actual localized output.