AI avatar video
An avatar video shows a generated person speaking to camera. Instead of a shoot day, a studio and a voice booking, the recording is made on a computer, in as many languages and versions as you need. We build avatar videos as a service, not as tool access: script, character, voice, image and edit arrive finished from us.
What an avatar video is
An avatar video is a talking head that was generated rather than filmed. The script comes first; voice, lip movement and image follow from it. The gain is not in the first video but in the second and the twentieth: if the text changes it is regenerated rather than reshot, and another language costs an audio track instead of a shoot day. The price for that is expression. An avatar speaks well; it does not act.
Avatar, digital twin or deepfake?
The words describe the same technique but very different things. What matters is whose face is on screen and whether that person agreed.
| Term | What it is | When it fits |
|---|---|---|
| Library avatar | a ready-made generated person from the provider's catalogue, licensed for advertising | fast and cheap, when the character need not be part of the brand |
| Brand avatar | a character designed for you alone | when the character should recur and be recognisable |
| Digital twin | the likeness of a real person from your company, your managing director for instance | when trust is tied to a real face, only with written consent |
| Deepfake | the likeness of a person without their consent | never. We do not make these, whoever orders them |
Avatar as a service, tool subscription and shoot compared
There are three routes to a talking head. The difference lies mostly in who does the work and what the tenth version costs.
| Route | Typical cost | Fits when |
|---|---|---|
| Run a tool subscription yourself | Synthesia lists 29 US dollars a month for the Starter plan at around twelve video minutes and 89 US dollars for Pro at around sixty minutes, plus your own working time | many short modules appear continuously and someone in-house builds them |
| Presenter and shoot | a shoot day with a small camera team usually 950 to 1,500 € net, every language version from scratch | a real face from your company should do the speaking |
| As a service with us | from 1,000 € net, further modules and languages far cheaper than the first | script, character, voice, image and edit should arrive finished |
The subscription prices come from the Synthesia pricing page, retrieved 24.10.2026, billed monthly. The shoot-day range comes from publicly listed prices of production companies in Austria and Germany, sources under Brand film cost.
Formats and lengths
An avatar video is almost never made on its own. So we fix format and length per place of use before the first module is produced.
| Where it runs | Format | Usual length |
|---|---|---|
| Training module in a learning portal | 16:9 | 2 to 4 minutes per module |
| Product explanation in a shop | 1:1 or 16:9 | 30 to 60 seconds |
| Internal announcement | 16:9 | 60 to 120 seconds |
| Social feed | 9:16 | 20 to 45 seconds |
What an avatar is good for
- Training and onboarding: twenty modules at a steady quality, without twenty shoot days.
- Product explanation in a shop: one person saying the essentials in thirty seconds.
- Multilingual communication: the same message in German, English and Russian, with matching lip movement.
- Content that changes often: prices, dates, terms. If the text changes it is regenerated, not reshot.
- Internal information: announcements and instructions nobody reads as an email.
Avatar video or explainer video?
Both explain, but in different ways. An avatar video relies on a person speaking and suits content that changes often or has to exist in many languages. An explainer video relies on images that show the content and is stronger when a process or a product has to become clear. If it is about onboarding and mandatory topics inside your own company, the training video is the right format, often with an avatar presenting inside it.
What we do not do
No avatars appearing as real customers recommending a product. Fake consumer endorsements are unfair under Annex Z 23c of the Austrian unfair competition act, however they were made. More on that on the page UGC videos with AI.
No likenesses of real people without their written consent, public figures included. And no voice that imitates a real person without the same consent.
Labelling
An avatar that looks and speaks like a human falls into the area for which Article 50 of the AI Act requires disclosure. We deliver every avatar video with the provenance note in the metadata, a ready line for the video description and, on request, a visible on-screen note. How that works in detail is in the guide to AI labelling in advertising.
Where AI reaches its limits
Avatars are strong when someone speaks calmly to camera. They are weak as soon as large gestures, real emotion or interaction with objects are needed. There the result quickly looks lifeless. In such cases we propose another route: an explainer video with a voice-over, or a short shoot with a real person that we combine with AI scenes.
How a project runs
- First call: 30 minutes, content, number of modules and languages.
- Script and character: you approve the text and the avatar, plus a voice sample.
- Test passage: about fifteen seconds to sign off tone, pace and effect.
- Production: the full version, then further modules and languages.
- Delivery: all versions with subtitles and a note on labelling.
That is the short version. All nine steps with the approvals are on Process.
Common questions
What does an AI avatar video cost?
Projects start at 1,000 € net. That covers script, character, voice, production and edit. Further modules and language versions cost considerably less, because character and voice are already settled. Anyone wanting to run the tools themselves pays from 29 US dollars a month at Synthesia by its own price list, but then carries the work.
Can the avatar look like our managing director?
Yes, as a digital twin, with her written consent. For that we need video footage and a voice recording as a basis. Without that consent we do not build a likeness of a real person.
Can you tell it is not a real person?
In calm speaking situations, often not any more. As soon as large gestures, movement through a room or interaction with objects come in, yes. We say openly in the first call whether your content suits an avatar.
How many languages are possible?
All the common ones, with matching lip movement. German and English are usual, plus whatever your markets need. We check the translation before production, because wrong text in an avatar video stands out particularly badly.
Does an avatar video have to be labelled?
As a rule yes. A generated person who looks and speaks like a human falls under the disclosure duty in Article 50 of the AI Act. We supply the note, in the metadata and as a ready line for the description.
For onboarding and mandatory topics inside your own company see training video, and for the line to faked customer voices the guide Testimonial video and AI.
Read on: Explainer video • UGC videos • Training video • Process