Lip sync is not simply a post-production convenience. It changes visible mouth motion to match a voice, potentially making a real person seem to utter words they did not say. That can be useful for authorized localization, but it can also become deceptive editing. The workflow therefore begins with purpose and consent, not with model settings.
Choose an authorized use case
An employee voluntarily localizing a training module, an actor approving a multilingual commercial and an instructional avatar based on its owner’s photo have identifiable authorization paths. An unconsented politician impression, altered witness statement or fake endorsement does not. The fact that an avatar provider allows an upload is not a release form. Establish rights to the image, voice, performance, script and finished distribution separately.
HeyGen requires a live consent video for a video-based Digital Twin and says the video must be recorded by the person represented. Synthesia documents a same-person live consent video for a Personal Avatar from photo. These are first-party safeguards, but a company should add its own purpose limitations and suspension process rather than treating one-time platform validation as unlimited permission.
Decide if changing the mouth is necessary
For an off-camera speaker or a narrated slide deck, audio-only dubbing may be sufficient and avoids facial edits. HeyGen publishes audio-only, Speed and Precision options with different charges and lip-sync behavior. Synthesia’s dubbing workflow lets operators toggle lip sync. Select the lowest intervention that communicates the content truthfully; a cinematic talking head is not inherently more accurate than good subtitles.
Conduct language and performance QC
Review translated words with a qualified speaker, checking names, numbers, idioms, dates, legal obligations and technical terms. Inspect facial artifacts at pauses, on plosive consonants, during side turns and when hands cross the face. Listen separately to detect clipping, unnatural pauses, room-tone shifts and background sound leaking through the new track. Preserve original audio so an auditor can compare material changes.
Label the result appropriately
A localized instruction video can say that its narration or lip movement has been generated when that is material to audience understanding. A documentary or interview should disclose alterations that may change the apparent testimony. Do not claim an avatar output is original raw camera footage. Keep consent records restricted, not publicly attached to an article or asset directory.
Final controls
The producer signs off on rights and purpose; a language specialist signs off on meaning; an editor signs off on synchronization; a subject representative approves any sensitive portrayal. Record who approved what and when. AI 05 retains ownership of standalone voice synthesis and music licensing research; AI 04 owns the video-specific consequences of merging them with a visible performance.
Worked production example
A real employee agrees to translate a training video but not to appear as an avatar in advertisements. Those permissions are different and should be recorded separately. Use an audio-only dub for passages where the employee is not visible; apply mouth synchronization only to approved on-camera segments. Have a native-language reviewer sign off on every changed safety instruction. If the localized delivery alters the apparent words without consent, block publication even if the lip sync is flawless. Store the source and dubbed masters privately with linked approval IDs.