What HeyGen is
HeyGen produces presenter-led video from text. You supply a script and an avatar; it produces a person speaking that script. It also translates existing video, replacing the voice and syncing mouth movement to the new language.
- Stock avatars, or a digital replica of a real person with consent.
- Text to speech in many languages and voices.
- Video translation with lip sync.
- Templates for common formats such as explainers and training.
What it is good at
| Use | Fit |
|---|---|
| Training and internal comms | Excellent |
| Product explainers | Excellent |
| Localising existing presenter video | Excellent |
| Sales and onboarding at scale | Strong |
| Cinematic or B-roll footage | Not applicable |
| Emotional or dramatic performance | Poor |
Choosing or creating an avatar
-
Decide stock or replica
Stock avatars are fast and carry no consent burden. A replica of a real person is more on-brand but requires written consent.
-
Match the avatar to the context
A corporate training avatar looks wrong in a consumer ad.
-
For a replica, record cleanly
Good lighting, neutral background, even pacing. The quality of the source recording sets the ceiling.
-
Get consent in writing
Specifically covering synthetic use, voice, platforms and term.
Voice is a separate clearance: clearing someone’s image does not automatically clear their voice. Get both explicitly.
Writing scripts that sound natural
- Write for the ear, not the eye. Sentences that read well silently often sound wrong aloud.
- Keep sentences short. Synthetic delivery degrades on long clauses.
- Read it aloud before generating — this catches most problems.
- Avoid numbers and jargon where possible, and check pronunciation when you cannot.
- Leave pauses. Natural speech is not continuous.
Generating and reviewing
- Generate a short section first and check voice quality before committing to the full script.
- Watch at full speed — stills hide timing problems.
- Check the first and last seconds of each section, where drift is most visible.
- Regenerate sections rather than the whole video when one part is wrong.
Translating video
-
Start from clean source footage
Mouth clearly visible, no occlusion.
-
Review the auto-translation
Machine translation of scripts needs checking, especially numbers and product terms.
-
Adjust length to fit
A translation that runs long causes rushed delivery and visible mismatch.
-
Check lip sync per section
Re-cut shots that will not sync rather than shipping a bad sync.
-
Keep the original audio track
So you can revert or offer both.
Consent and disclosure
- Written consent naming the specific use, platforms and term for any replica.
- Separate consent for voice cloning.
- Disclosure to audiences where required — several jurisdictions regulate synthetic performance.
- Retain consent records with the project files.