The most common mistakes with AI avatars in HeyGen and Synthesia are using them where viewers expect a real person, writing scripts for reading rather than speaking, skipping consent and disclosure, and localising without native review. Avatar tools turn a script into a video with a synthetic presenter in many languages. They save real production time for training, explainer and update videos when used for the right jobs.
Both companies offer stock avatars, custom avatars built from footage of a real person, voice cloning and translation. Below are the mistakes we see most, and how to avoid them.
Mistake 1: Using avatars where trust depends on a real person
Testimonials, founder messages, apologies and anything emotional need a real human. An avatar delivering a customer story reads as fake, and in many markets it would mislead viewers.
Avatars fit informational content: onboarding, product updates, internal training, how-to guides, FAQ answers and multilingual versions of existing explainers.
Mistake 2: Writing for the page, not the voice
Scripts pasted from blog posts produce stiff videos. Long sentences, nested clauses and lists of numbers sound unnatural when spoken by any presenter, synthetic or not.
- Keep sentences short and conversational
- Write numbers the way they should be said
- Add pauses with line breaks or the tool’s pause markers
- Read the script aloud before generating; if you stumble, the avatar will sound odd too
- Check pronunciation of brand and product names and use the phonetic override features
Mistake 3: Cloning someone without proper consent
Custom avatars and voice clones are built from a real person’s likeness. Both HeyGen and Synthesia require consent from the person, usually through a recorded consent statement. Beyond the platform rules, get written agreement covering how the avatar may be used, for how long, and what happens when the person leaves the company.
An avatar of a former employee still presenting your product videos is an awkward problem to discover after the fact.
Mistake 4: No disclosure
Viewers increasingly recognise AI presenters, and many platforms require labelling of realistic synthetic media. Disclose in the video description or with a small on-screen note. Being open costs little; being found out costs trust.
Why do localised avatar videos often fail?
Translation and lip-sync features can produce versions in dozens of languages quickly. The output can still contain wrong terms, unnatural phrasing or culturally odd examples. A Turkish explainer translated to German may keep a joke or a reference that makes no sense there.
- Translate the script first and have a native speaker edit it
- Generate the video from the edited script, not from automatic translation of the video
- Check on-screen text and graphics separately; they are not always translated
- Review lip-sync on close-up shots
Mistake 5: Static, slide-like videos
A talking head in front of a plain background for three minutes loses viewers. Use the tools’ scene features: change layouts, add screen recordings, product images and text overlays. Keep videos short and break long topics into a series.
Mistake 6: Forgetting to measure
Production speed is easy to see. Whether viewers watch is not. Track view duration in your video host, completion rates in your LMS for training content, and support ticket volume after help videos go live. If watch time drops sharply after the first seconds, the opening or the format needs work.
How Alien Road applies AI avatars
In a typical engagement, our team first sorts the client’s video needs by trust level: content that must feature real people and content that can use avatars. Only the second group moves to HeyGen or Synthesia, chosen by the client’s language needs, security requirements and existing tools.
We write scripts for speech, set up pronunciation for product names, and when a custom avatar is used, handle the consent and usage agreement with the person involved. Every language version is edited by a native speaker before generation. Reporting tracks production time per video, watch time and completion, and for help content, changes in related support questions.
HeyGen or Synthesia: how do you choose?
Both offer large stock avatar libraries, custom avatars, voice options, translation and templates. Differences change with each release, so compare current features against your needs rather than reviews from a year ago.
- Language list and voice quality in the specific languages you need
- Custom avatar process: footage required, consent workflow, turnaround
- Collaboration features: shared workspaces, brand kits, approval steps
- Security and compliance: SSO, data retention, certifications your IT team requires
- Integrations with your LMS, video host or design tools
- Pricing model: minutes per month, seats, credits
Run the same two scripts in both tools on trial accounts and have the final viewers, such as customers or staff, compare them without knowing which is which.
Mistake 7: Ignoring the update cost
One advantage of avatar video is that edits are cheap: change a line in the script and regenerate. Teams often forget this and treat videos as fixed. Product changes then leave old videos on help pages for months. Keep scripts in a shared document with an owner and a review date, so updating a video is a ten-minute task rather than a forgotten one.
Mistake 8: One video for every channel
A 16:9 training video does not work as a vertical social clip, and a social clip is too short for onboarding. Plan formats per channel from the script stage. Most tools can output different aspect ratios from the same project, but the script length and pacing should change too.
Where to start this week
Pick one existing written how-to guide that customers often ask about. Rewrite its first section as a 60-second spoken script, read it aloud, then generate it with a stock avatar in your current language. Show it to three customers or colleagues and ask whether they would watch the rest. Their answer tells you more than any feature list.
Keep a list of every avatar video in use, with its script owner and review date, so updates happen on schedule.
Keep reading: Video Production · AI Marketing