AI avatar platforms such as HeyGen and Synthesia turn a written script into a video presented by a synthetic or cloned human presenter, with lip-synced speech in many languages. A good workflow runs from script to avatar and voice selection, scene building, translation, review and publishing. The time saving comes from replacing filming and reshoots for training, explainer and update videos, not from replacing every kind of video.
Both platforms work in the browser. Synthesia has focused heavily on corporate training and internal communication, with templates, brand kits and integration into learning platforms. HeyGen is popular with marketers for personalised and social video, translation with voice cloning and its interactive avatar features. Feature sets overlap and change often, so check current plans before choosing.
Which videos suit AI avatars?
Avatars work best when the message matters more than the performance. Viewers accept a synthetic presenter for information; they are less forgiving when the video is meant to show emotion, real people or a real place.
- Product walkthroughs and feature updates that change every few months.
- Onboarding and training modules for staff or customers.
- Localised versions of an existing video in several languages.
- Short internal announcements and policy explainers.
- Personalised sales follow-ups at scale, with clear disclosure.
Poor fits include brand films, customer testimonials and founder messages presented as if filmed live. A testimonial voiced by an avatar is not a testimonial, and presenting it as one misleads viewers.
What does the step-by-step workflow look like?
Treat avatar video like a small production with fewer moving parts. Each step has an owner and a short checklist.
- Script: write for the ear, with short sentences and spoken numbers; aim for roughly 130 to 150 words per minute of video.
- Avatar and voice: choose a stock avatar or a consented custom avatar, and a voice that fits the audience and language.
- Scenes: add slides, screen recordings, brand colours and on-screen text through the platform’s brand kit.
- Pronunciation: set custom pronunciations for product and brand names.
- Review: watch the full video with sound, checking lip sync, names, numbers and on-screen text.
- Translate: generate language versions, then have a native speaker review each one.
- Publish and archive: export, add captions, store the script and project link together.
The pronunciation step saves the most rework. Brand names, Turkish place names and technical terms are often read wrongly by default. Both platforms offer ways to adjust pronunciation; set them once and reuse them across projects.
Updating instead of reshooting
The biggest gain arrives months later. When a product interface changes, you edit the script and swap the screen recording, then regenerate. A filmed video would need the presenter, the set and an editor again. Keep master projects organised so updates take minutes.
How should consent and disclosure be handled?
Custom avatars of real people require that person’s explicit consent. HeyGen and Synthesia both build consent steps into custom avatar creation, typically a recorded statement. Keep a written agreement too, stating where the avatar may be used and what happens when the person leaves the company.
Disclose synthetic presenters where viewers could be misled. YouTube asks creators to label realistic altered or synthetic content, and Meta applies AI labels in some cases. The EU AI Act also includes transparency obligations for deepfake-style content. A short line in the description or on screen is usually enough and avoids awkward questions later.
How do you keep quality consistent across a library?
Use templates. Both platforms let you save scene templates with brand fonts, colours, logo placement and intro and outro screens. A library built on three or four templates looks coherent even when many people produce videos.
Limit the avatar roster. Picking a new avatar for every video makes the library feel random. Choose one or two presenters per audience and keep them.
Review translations with people, not only tools. Machine translation and voice cloning are good at the time of writing but still produce awkward phrasing and occasional mistakes in numbers and names. A native reviewer watching each language version is a small cost against a public error.
How do you measure the time saved?
Record hours per finished minute of video before and after. Include scripting, filming or generation, editing, review and translation. For localisation, compare the cost of producing a new language version by dubbing or reshooting with generating it.
Measure viewer response as well. Watch-through rates in your video host or learning platform show whether audiences stay with avatar videos. If completion drops compared with filmed versions, shorten scripts or mix in more screen content.
Which roles does the team need?
A small avatar video operation needs three roles, which can sit with two people. A script writer who understands spoken language and the subject. A producer who builds scenes, handles pronunciation and exports. And a reviewer, ideally from the audience side, such as a trainer or product manager, who checks accuracy.
For localisation, add a native reviewer per language. Freelance reviewers work well here because the task is short and specific: watch, note errors with timestamps, approve. Agree a turnaround time so translation does not become the bottleneck that cancels out the speed gained elsewhere.
How Alien Road applies AI avatars
In a typical engagement, discovery lists the client’s recurring video needs and sorts them into avatar-suitable and film-only. We also check who would appear as a custom avatar, if anyone, and draft the consent agreement with the client.
Setup means choosing the platform that fits the use case, building brand templates, setting pronunciation dictionaries for brand and product names, and writing a script style guide. Each video goes through the workflow above, with native review for every language version.
We measure production hours per finished minute and watch-through rates, and report both monthly. When a video type performs worse as an avatar version, we recommend filming it instead.
Where to start this week
Pick one existing training or explainer video that needs updating. Rewrite its script for spoken delivery, then produce an avatar version on a free or trial plan of HeyGen or Synthesia.
Show both versions to five people from the target audience and ask which they would rather watch to learn the same thing. Their answers, plus the time you spent, will tell you whether avatar video belongs in your workflow.
Keep reading: Video Production · AI Marketing