What is an AI digital twin, and do you need one?
Two industries use the same phrase. Engineers mean a live software model of a machine. We mean a person: a saved version of a real face and a real voice that can speak whatever you write, and only while that person's consent is on file. This page sorts out which one you came for, then helps you pick between a Digital Twin and an Avatar.
Two things share the name. Only one of them is a person.
The glossary defines the words and the build page walks the steps. This page helps you choose between them.
Part of Resources.
Checked September 2026. Free to read, and there is nothing to sign up for here.
What is an AI digital twin?
An AI digital twin is a saved version of a real person, built from their photo and voice, that can speak a script on screen. In engineering the same phrase means a live model of a machine. On Kyndrify it means the person: we call it a Digital Twin, and it only works with that person's recorded consent.
Who it is for
Anyone working out whether they need a version of a real person on screen, a generated character, or neither.
What you need
For a Digital Twin: one clear photo of the person, about fifteen seconds of them on camera agreeing, and a voice. For an Avatar: nothing of yours at all.
What it costs
Reading this costs nothing. A paid Digital Twin video is quoted at 152 credits a second. The Free plan includes 8,400 credits once and holds a talking-head video to 60 seconds. Full-length videos in your own twin start on the Plus plan, $49 a month billed yearly, with 28,000 credits a month.
What it does not do
It is not built from a video of you talking. It cannot be made of somebody whose own recorded consent you do not hold. And an Avatar is not a person, so it must never be shown as one.
One phrase, two industries.
In engineering, a digital twin is a software model of a real physical thing that keeps up with the real thing: a turbine, a production line, a building. Sensors feed it, and people run tests on the model instead of on the machine. That is a serious field and it has nothing to do with video. If that is what you came for, you are in the wrong shop, and the rest of this page will not help you.
In video, the phrase means a person. A saved version of someone's face and voice that can present a script without them filming it again. Most tools call this an AI twin, and we call it a Digital Twin, which is the word you will see everywhere in our Studio.
The two meanings never meet. Nothing on this page simulates anything. There is no model of you that predicts what you will do. There is a saved likeness that reads what you wrote, and a consent record that says you agreed to it.
A real person, or a character we made.
This is the choice that actually matters, and most pages blur it. A Digital Twin is somebody. An Avatar is nobody. Everything else follows from that one line.
| What changes | Digital Twin | Avatar |
|---|---|---|
| Who is on screen | Digital TwinA real person. You, or someone who agreed on camera. | AvatarA character we generated. Nobody real. |
| What it needs first | Digital TwinA clear photo, a recorded consent, and a face in that recording that matches the photo. | AvatarNothing. Pick one and start. |
| Does permission expire | Digital TwinYes. Consent runs out after a year, and the twin stops rendering until it is recorded again. | AvatarThere is no consent, so there is nothing to expire. |
| Can it be taken back | Digital TwinYes. The person withdraws it themselves, and after that no new videos of them can be made. | AvatarNothing to withdraw. |
| On the free plan | Digital Twinone Digital Twin for your Workspace, for life. | AvatarYou use one of ours. Building your own Avatar starts on the Plus plan. |
| Good for | Digital TwinAnything where the face has to be yours: your channel, your firm, your pitch, your own clients. | AvatarAnything where the face just has to be a face: an explainer, a how-to, a product walk-through. |
| Cannot do | Digital TwinBe made of a person who did not say yes on camera. | AvatarStand in for a real customer, or carry a real person's word. |
The consent test: three questions, one answer.
Before we render anything with a face, the product asks the same three questions you should. Answer them about your own situation and the choice makes itself.
Is there a real person in this?
Not does it look like a person. Is the video meant to be a particular human being, someone a viewer could name? If yes, you are in Digital Twin territory, and the next two questions are not optional.
Do you hold that person's yes, on the record?
A text message is not a record. He said it was fine is not a record. A record is that person, on camera, saying they agree, with the face in the recording matching the face in the photo. If you do not have it you can still get it: send them a private link and they record it themselves, on their own phone, in about a minute.
Does the face have to be theirs?
Sometimes the honest answer is no. A product explainer, a how-to, an internal update: those work with any believable presenter. If the message does not depend on whose face it is, an Avatar is faster, nobody's likeness is involved, and there is nothing to keep current.
The verdict. Yes, yes, yes means build a Digital Twin. Yes, no, yes means get the consent first and build nothing until you have it. A no on question three means pick an Avatar and get on with your day.
What this looks like on a Tuesday.
The producer with a client and no release.
She makes weekly videos for a founder who has no time to film. Question one, yes, it has to be him. Question two, no, all she has is an email thread. So she sends him a consent link, he records it on his phone between meetings, and the twin becomes hers to render from then on. If he ever wants out, he withdraws it himself and she cannot make new videos of him. That is the paperwork, and it took him about a minute.
The course creator who is in module four and not module five.
She appears in the modules where she is teaching, so those use her Digital Twin. Module five is a software walk-through where nobody needs to see her at all. Question three, no, the face does not have to be hers. She picks a Featured character and renders it the same afternoon. No consent to collect, because there is no person to ask.
What it costs, and which door opens on which plan.
Credits are charged by the second, so the cost follows the length of the video rather than the plan you are on. A paid Digital Twin video is quoted at 152 credits a second.
On the Free plan you get 8,400 credits once, one Digital Twin for your Workspace for life, and a talking-head video that runs to 60 seconds. Every Featured character is open to you too, so you can make an Avatar video on day one without building anything. What is not included: building your own Avatar, and full-length videos in your own twin. Both start on the Plus plan, $49 a month billed yearly, with 28,000 credits a month.
That is the whole shape of it. The full ladder is on Pricing, the step-by-step build is on Create an AI version of yourself, and what a finished video looks like is on AI talking head generator.
Six things not to say about a video like this.
The technology is fine. The description is where people get into trouble.
Do not call an Avatar a customer.
A generated character saying nice things about your product is an ad, not a testimonial, and dressing it as one is the fastest way to lose an audience.
Do not say it was filmed.
It was rendered. Nobody minds, as long as you do not pretend otherwise.
Do not call a Digital Twin a clone of the person.
A voice clone is a real thing, and those are the right words for what the voice is. The person is not copied. A Digital Twin is a saved likeness that reads what you wrote, and it stops there.
Do not say someone consented when you only asked them.
Their yes lives in a recording they made themselves, or it does not exist.
Do not promise that a mark proves it came from you.
Free renders carry a small visible mark, and paid downloads carry none. Built-in labelling is still rolling out across output types, so do not build a policy on it yet.
Do not say digital twin to an engineer without saying which one.
They will picture a turbine, and you will spend ten minutes climbing back out.
The part that is yours, not ours.
Rules about likeness, biometric information and disclosing synthetic media differ by country and by state, and some professions have their own advertising rules on top. Kyndrify gives you the consent record, the face check and the marks on the file. It does not make your video legal where you publish it, and we do not claim that using it satisfies any law or rule. Where a disclosure is required, the duty sits with the person publishing the video. If your field is regulated, check with whoever normally checks your advertising before you post.
What we hold when you build a Digital Twin, and how long we hold it, is written out in the Biometric information privacy notice. How we handle consent, disclosure and content credentials is on Responsible AI.
Checked September 2026. This is a summary, not legal advice.
Questions people ask about the two kinds.
- Is an AI digital twin the same as a deepfake?
- The technology overlaps. The difference is permission and a record. A Digital Twin needs the person's own recorded consent before it can render anything, we compare the face in that recording with the face in the photo before we accept it, and the person can withdraw it later. A deepfake is the same idea with the consent missing. That is the whole distinction, and it is not a small one.
- Can I make one of a colleague, a client or someone famous?
- Only with their own recorded consent, recorded by them. You cannot record it on their behalf, and there is no path where you upload a photo of someone and get a twin of them. Send them a private link, they do it themselves, and they can take it back whenever they want.
- Is a digital twin the same thing as an AI avatar?
- No, and the table above is the short version. A Digital Twin is a real person and carries consent. An Avatar is a generated character and carries none. The word avatar gets used for both by other tools, which is most of why this page exists.
- What happens to my face after I build one?
- We keep one encrypted template of it, so we can check that a new photo or a new recording is still you. The readings taken for each individual check are not kept, and we do not use any of it to train a Kyndrify model. The consent recording is held as evidence for at least the life of the twin plus seven years, because that is what makes it evidence. Deleted Digital Twin data can stay recoverable for up to 30 days, and is then hard-deleted.
- Do I need a Digital Twin to make a video at all?
- No. Pick a Featured character and you can render something today with nothing of your own in it: every Featured Avatar is free to use in a render on every plan, the Free plan included. Plenty of people start there, decide they like the format, and build their own twin a week later.
- Does the engineering meaning matter to me?
- Almost certainly not. If you arrived here wanting to model a machine or a building, this is the wrong kind of twin and the rest of the site will not help. If you arrived wanting to be on video without filming, you are in the right place.
Read next
Now pick the one you actually need.
If the face has to be yours, build a Digital Twin: one photo, about fifteen seconds of you saying yes, and your consent on file. If it just has to be a face, start with one of ours.