Feature
How AI avatars work: a face and a voice become a talking video
An AI avatar is a photo or a short video of a person that an AI model animates to match a voice recording. The model moves the mouth, jaw and head in time with the sounds.
In short
- An AI avatar is built from one photo or a video of a person, then animated so the mouth matches a voice.
- The voice can be your own recording, text-to-speech made from your script, or in some tools a clone of your voice.
- The mouth movements come from the sound, so clean speech with no echo or music gives the best lip sync.
- Blurry teeth, odd blinking and a head that moves while the shoulders stay still are common signs of an avatar.
- In Cutroom, avatar video costs 350 credits a minute, plus 70 a minute if Cutroom generates the voice.
- Photo is all a photo avatar needs
- 1
- Voice-over length one Cutroom avatar accepts
- 3 s to 5 min
- Credits per minute of avatar video in Cutroom
- 350
The two kinds of AI avatar
AI avatars are made in one of two ways. Some tools animate one photo, and others build the avatar from a video of a real person.
- Stock avatars are often built from real actors who licensed their face. TikTok, for example, says its Symphony avatars are licensed actors.
- Cutroom has ready-made stock presenters too, which you can pick instead of your own photo.
Photo avatars, video avatars and stock avatars
Tool details from each company's own site, checked 21 September 2026.
| Kind | Made from | What moves | Examples |
|---|---|---|---|
| Photo avatar | One photo of a face | The mouth, jaw, face and head | D-ID, Cutroom |
| Video avatar | A video recording of a real person | The face and mouth, over body movement from the recording | HeyGen's digital twin, Synthesia's personal avatars |
| Stock avatar | A ready-made person the tool offers | Depends on whether it was made from a photo or a video | HeyGen 500+ on the free plan, Synthesia 125+ on Starter |
Kind: Photo avatar
- Made from
- One photo of a face
- What moves
- The mouth, jaw, face and head
- Examples
- D-ID, Cutroom
Kind: Video avatar
- Made from
- A video recording of a real person
- What moves
- The face and mouth, over body movement from the recording
- Examples
- HeyGen's digital twin, Synthesia's personal avatars
Kind: Stock avatar
- Made from
- A ready-made person the tool offers
- What moves
- Depends on whether it was made from a photo or a video
- Examples
- HeyGen 500+ on the free plan, Synthesia 125+ on Starter
How a photo avatar is made, step by step
Photo avatar tools work in roughly the same way. You give them a face and a voice, and you get back a video of that face talking.
- The AI model draws new video frames from your photo. The mouth, jaw and head move, and the rest of the picture stays almost still.
- The mouth shapes come from the sound of the voice, not from the written script.
- Text-to-speech turns a typed script into a spoken voice, so you do not have to record anything.
- Voice cloning copies a real person's voice from a recorded sample. HeyGen offers it from its Creator plan, and Cutroom does not clone voices.
- Cutroom makes one continuous video that is exactly as long as your audio, from 3 seconds to 5 minutes.
How a photo becomes a talking video
A photo
A front-facing picture of a face
A voice
A recording, or a voice generated from text
The AI model
Listens to the sound and moves the face to match
The video
The face says your words
This is the whole editor
Highlight a phrase and a clip lands on those exact words. No timeline, no keyframes, no layers.
Signs that give an AI avatar away
Most of these signs have the same cause. The AI works from one picture and a sound, and it has to guess the rest.
What viewers notice, and why it happens
| What viewers notice | Why it happens |
|---|---|
| Teeth that blur or change shape | A closed mouth in the photo hides the teeth, so the AI has to invent them in every frame. |
| Blinking that looks too regular, or eyes that stare | The voice tells the AI how the mouth moves, not where to look or when to blink. |
| A head that moves while the shoulders stay still | A photo avatar mostly animates the face and head. The body stays as it was in the photo. |
| Lips slightly off the words | Echo, music or a muddy recording hides the sounds the mouth shapes come from. |
| A face that does not match the words | The face follows the sound, not the meaning, so it does not laugh or frown on cue. |
What viewers notice: Teeth that blur or change shape
- Why it happens
- A closed mouth in the photo hides the teeth, so the AI has to invent them in every frame.
What viewers notice: Blinking that looks too regular, or eyes that stare
- Why it happens
- The voice tells the AI how the mouth moves, not where to look or when to blink.
What viewers notice: A head that moves while the shoulders stay still
- Why it happens
- A photo avatar mostly animates the face and head. The body stays as it was in the photo.
What viewers notice: Lips slightly off the words
- Why it happens
- Echo, music or a muddy recording hides the sounds the mouth shapes come from.
What viewers notice: A face that does not match the words
- Why it happens
- The face follows the sound, not the meaning, so it does not laugh or frown on cue.
Why the audio matters so much
The model reads the mouth shapes from the sound. If the words are unclear in the audio, the mouth looks unclear in the video.
- Echo from hard walls blurs the words together, so the mouth looks mushy.
- Music or background noise under the voice confuses the model. Upload speech only and add music later.
- A low-quality audio file loses sounds like p, b and m, where the lips close, and the lip sync gets vaguer.
- Record in a room with carpet and curtains. Soft surfaces cut echo, even when you record on a phone.
- A generated voice has no room sound at all, which is why it usually lip-syncs cleanly.
What makes a good avatar photo
The photo sets the look of the whole video, and some photos animate badly.
- Face the camera straight on, with your shoulders in the frame.
- Use soft, even light, like a window, and a plain wall behind you.
- Keep your mouth closed or relaxed. A wide smile often looks strange once the face starts talking.
- Avoid portrait-mode blur, hats and high collars near the jaw. They smear or look stuck once the head moves.
- Cutroom accepts PNG, JPEG or WebP photos up to 15 MB.
How to write a script an avatar can say well
A generated voice keeps an even tone, and an avatar face shows little expression. A script written for reading sounds stiff when it is spoken that way.
- Short sentences with one idea each sound the most natural.
- About 100 words fill 40 seconds at a normal speaking pace of 150 words a minute.
- Personal lines like "it changed my mornings" sound false from an avatar. Film those yourself.
- Read the script out loud once before you make the video, and rewrite any sentence you stumble on.
What a photo avatar cannot do
Knowing the limits tells you which videos to make with an avatar and which to film.
- It cannot pick up, open or point at a product, because it is made from a still photo.
- It stays in the same room, at the same distance from the camera, for the whole video.
- Its face does not react the way a real person does, such as laughing, frowning or looking away.
- One Cutroom avatar video takes up to 5 minutes of audio.
- In Cutroom you can use your own face, one of the stock presenters, or the face of someone who gave you written permission.
Where Cutroom fits
Cutroom turns your photo and voice into an avatar video, then edits it into a 9:16 ad with captions and product clips. Avatar video is 350 credits a minute.
- With your own recorded voice, the same ad costs about 348 credits.
- A 90-second avatar video costs 525 credits for the avatar alone, more than twice the 40-second one.
- The avatar video is not ready instantly. Cutroom shows how long it will take and emails you when it is ready.
- Avatars start on Basic, $39.99 a month for 2,500 credits, where a finished 40-second avatar ad works out at about $6.30.
A 40-second avatar ad in Cutroom, step by step
Each charge is rounded up to a whole credit.
| Step | Price | For 40 seconds |
|---|---|---|
| Generated voice-over | 70 credits a minute | About 47 |
| Avatar video | 350 credits a minute | About 234 |
| Edit: cuts, captions, extra clips | 100 credits | 100 |
| Export | 20 credits a minute | About 14 |
| Total for a 40-second ad | Voice, video, edit and export | About 395 |
Step: Generated voice-over
- Price
- 70 credits a minute
- For 40 seconds
- About 47
Step: Avatar video
- Price
- 350 credits a minute
- For 40 seconds
- About 234
Step: Edit: cuts, captions, extra clips
- Price
- 100 credits
- For 40 seconds
- 100
Step: Export
- Price
- 20 credits a minute
- For 40 seconds
- About 14
Step: Total for a 40-second ad
- Price
- Voice, video, edit and export
- For 40 seconds
- About 395
Questions people ask
- What is the difference between a photo avatar and a video avatar?
- A photo avatar is animated from one picture, so mainly the face and head move. A video avatar, like HeyGen's digital twin, is built from a recording of a real person, so it starts from real movement.
- Does the AI animate the whole photo or only the mouth?
- Mostly the mouth, jaw and face, with small head movements. The background stays almost still, which is why a hat brim or high collar near the jaw can look stuck.
- Why does my avatar look worse than the demo videos?
- Usually the audio or the script is the problem. Echo in the recording blurs the lip sync, and a script written for reading sounds stiff when it is spoken. Demo videos use clean studio speech.
- Do I need to record my own voice?
- No. Cutroom can generate a voice from your script for 70 credits a minute. Your own recording is free and sounds like you, as long as the room has no echo.
- Can I fix an avatar video after it is made?
- Not the avatar video itself, but you can fix the ad around it. Add captions, fix a misheard word and cover parts with other clips in the Cutroom editor, for 100 credits. A line you want gone needs new audio and a new avatar video.
- How much does an AI avatar video cost?
- In Cutroom, 350 credits per minute of avatar video. A finished 40-second ad with a generated voice, edited and exported, comes to about 395 credits.
- Can I make an avatar of someone else?
- Only with their written permission. Cutroom only allows your own face, a stock presenter, or a face whose owner agreed in writing.
Start with clean audio, a plain photo and a short script. Avatars come with Basic at $39.99 a month, and your first filmed video is free.