A software trainer recorded an English-language video explaining how to reset a password, and a partner needs a Spanish version. A fluent translation matters more than a convincing moving mouth: a wrong instruction could lock users out. HeyGen offers video translation, optional lip-sync processing and review tools, but it also documents restrictions on Proofread and on-screen text. This guide gives the trainer a glossary, a translation approval table and a frame-based lip-sync review. The presenter drawings are original examples, not HeyGen output.
Start with a source script that is actually approved
Transcribe the source video and check every instruction against the real application. Mark product names, user interface commands, URLs and support contacts. A translator cannot reliably repair an inaccurate English source; fix wrong steps before generating another language. Record consent from the speaker for using a likeness and voice in translation, and confirm the video is yours to process. Write a list of words that must remain unchanged, such as a brand name, alongside phrases that should be localized naturally. This source script is the stable reference for reviewers, not a model-produced translation treated as ground truth.
Verify what the chosen HeyGen mode does
The official HeyGen Video Translation guide describes translation and available lip-sync modes as well as plan-dependent review tools. It distinguishes Speed and more controlled modes and notes that the best results need a clearly visible speaker, steady footage and clean speech. Choose a mode appropriate to the source instead of promising that lip movements will exactly match every phoneme. Check access in the account before starting; plan availability and pricing can change. A talking-head help tutorial with one visible speaker is easier to inspect than footage with frequent cuts or several overlapping voices.
Build a terminology glossary before translation
Prepare a two-column document: approved English term and preferred Spanish rendering, plus short context notes. HeyGen calls its terminology aid Brand Glossary and explains how it is used. Include commands such as “Reset password,” identifiers that should not be translated, and the difference between “sign in” and “sign up.” Have a competent Spanish reviewer approve the glossary, not merely a monolingual editor comparing machine outputs. For product instructions, verify that the target-language app actually uses the chosen translated UI label. Otherwise a natural-sounding video may send viewers to a button they cannot find.
Use Proofread deliberately rather than assuming an SRT locks words
HeyGen describes Proofread, also called Review & Edit, as a way to inspect the translated script before the final video; availability depends on plan type. Its guidance makes an important distinction: an input SRT uploaded before translation guides the process but is retranslated, whereas an output SRT supplied in Proofread mode is spoken as written. Do not tell a team that any SRT upload freezes the target transcript. Record which path you used and save the final approved target-language text. Recheck timing after editing longer Spanish sentences.
Inspect changes in duration and mouth movement
HeyGen documents Dynamic Duration that can stretch or compress segment length within specified limits and warns that lip sync quality depends on speaker positioning. Watch target audio and face movements together at real speed, especially at the start of each sentence and after pauses. A translation may be semantically correct but visually unnatural because a word falls during a closed mouth. Review short sections around names and numbers rather than relying on a single first-frame screenshot. If the face twists or teeth are distorted, flag the segment for revision or use a less aggressive mode rather than claiming perfect synchronization.
Separate spoken translation from visible screen text
The HeyGen Help Center explicitly says Video Translation does not translate on-screen graphics, built-in captions or interface text. For the software tutorial, a Spanish voice over an English-only menu can create confusion even when spoken instructions are correct. Make a separate graphics inventory: every screen label, title card, lower third, embedded subtitle and URL. Decide which elements need a localized remake in the original video editor. Do not quietly imply that auto lip sync also localized the application interface. Keep the exported audio and on-screen graphics under the same language review process.
Approve meaning, pronunciation and accessibility individually
Ask a qualified native-language reviewer to assess password steps, account names, numerical values, tone and pronunciation. Next, review subtitle readability and timing; a dubbed voice alone may not meet accessibility needs. Finally check likeness, lip sync and whether the source speaker appears to make a claim they never approved. Use a ledger with columns for segment, source meaning, target text, visible issue, reviewer and release decision. Mark missing approvals as pending. These are practical review categories, not a measured model-performance study. If any key instruction is wrong, withhold that version until corrected.
Package the localized asset with an approval trail
Archive the original approved recording, source transcript, translation glossary, target transcript, Proofread decisions, consent record, generated video and final captions. Keep edits to visible UI and captions as separate assets so the team can update them when software changes. Verify the actual exported file on phone and desktop, paying particular attention to lip timing after download. Label any synthetic translated voice honestly where context or platform policies call for it. The production goal is an understandable and authorized Spanish tutorial, not an automatic equivalence claim between translation software and professional review.
How We Research
On October 28, 2026, we consulted HeyGen’s official Video Translation guide for Proofread, glossary, SRT distinctions, Dynamic Duration and the on-screen-text limitation. We did not translate a video or measure lip-sync quality; all interface and presenter drawings in this article are conceptual. A qualified language reviewer and consent checks are required for real publication.
Frequently Asked Questions
No. The help guide says input SRT is retranslated; a Proofread output SRT is handled differently.
No. HeyGen says on-screen text and graphics must be localized separately.
Yes. Especially when names, numbers and functional instructions matter.
Yes. Review meaning and facial motion as separate checkpoints.
Source script, glossary, approved target text, consent, generated clip, captions and review decisions.