Can AI create videos of yourself?

Yes, and this is how: you give a photo, a short recording where you consent out loud, and a voice. After that you type a script and the video comes back in your own face. There is no filming and no studio, and nothing is made that you did not write.

Do I need a video of myself to build a twin?

No, and you cannot use one. A twin is built from a still photo plus a short consent recording. If you have a video of yourself talking, it is not the input this takes.

How long does the consent recording have to be?

From fifteen to ninety seconds. Fifteen is genuinely enough. Anything under fifteen comes back and asks for a few more seconds, and anything over ninety asks you to keep it shorter.

What if the face check says no?

Record again in the same light and at the same angle as your photo. The check is set to be cautious on purpose, so it would rather ask twice than let a photo of somebody else through. It is not a mark against you, and recording, or re-recording, a consent clip costs no credits.

Can I build a twin of somebody else?

Only with their own recorded consent, given by them. You send them a private link, they record on their own device, and they can withdraw it later. You cannot record it for them, and a twin built from a photo of a person who has not consented is one we take down.

What happens to my photo and my recording?

A consent recording is kept for at least the life of the twin plus seven years, because it is the evidence the person agreed. One encrypted face template is kept per person, and the readings taken for each check are not kept. Kyndrify does not use your photos, voice, scripts or finished media to train a Kyndrify model. Deleted Digital Twin data can stay recoverable for up to 30 days, and is then hard-deleted.

Resources · Digital Twin

How to create an AI twin, step by step

One photo of your face, a fifteen second recording where you say yes out loud, a look, and a voice. Here is what each step asks for, in the order it asks, and what happens when one of them sends you back.

One photo. A fifteen second recording. A voice. Here is every step.

This walks through what the builder asks for today. It is not legal advice, and the rules about recording a person differ by where you and your viewers are.

Part of Resources.

Checked September 2026. Read end to end. Free to read, with no sign-up.

A man in a grey sweater talks to camera over a coffee at his kitchen counter.

How do you create an AI twin of yourself?

You build one from a photo and a short recording, not from a video of you talking. The builder asks for a clear photo of your face, a fifteen second recording where you say yes and read a code we put on screen, a look, and a voice. We check the two faces match, then your twin is ready.

Who it is for
Anyone who has to be on camera more often than they can film. You have to be the person in the photo, or hold that person's own recorded consent.
What you need
One clear photo with just your face in it, a quiet spot, a camera you can talk into for fifteen seconds, and a name for the twin.
What it costs
Reading this costs nothing. The photo, the recording and the first sets of looks are not charged for. A finished video in your own twin is 152 credits a second on the Plus plan and above. Free starts you with 8,400 credits once, and caps one video at 60 seconds. See the plans.
What it does not do
It will not build a twin from a video of you talking, it will not let you record consent for somebody else, and on Free the first run ends on a four second sample rather than a finished video.

Three things to have ready, and one to decide

Have these to hand and the whole thing runs without a pause.

A photo, or a camera and thirty seconds of privacy.
You can upload a photo you already have, or let the guided capture take one. The capture draws an outline to sit inside, gives you one line of direction, counts you in so nobody is caught mid-blink, and lets you retake as often as you like.
Somewhere you can talk out loud.
The recording has words in it, and you read them off the screen. A room with a door is enough. Headphones are not needed.
A name for the twin.
Yours is fine. If you are building several, name them for the person rather than the project, because that is the name you will be picking from later.

Digital Twin: The version of a real person, built with their consent, and the thing this page walks you through.
AI avatar: A character you pick from a library or design from nothing. Nobody consents to one, because nobody is in it.

The one thing to decide before you start is whose twin this is. If it is you, you record the consent yourself. If it is somebody else, they record it, on their own device, through a private link you send them. You cannot record it on their behalf, and that is the point of it.

The build order: five steps, and what each one decides

The paid builder asks in this order. Every step decides one thing, and knowing which thing is what stops people redoing the wrong step later.

Photo. This decides the face.
Everything the twin ever makes is a picture of this picture, so this is the step worth being fussy about. Look straight into the lens. One person in frame. Plain light, and nothing behind you that you would not want in a video.
Record. This decides the permission, and the way you move.
You read a short script off the screen that says you are building your twin and you agree to it being used in the videos you make. Partway through there is a code we put on the screen for you to read out. The take is consent first. It is also read for the way you carry yourself, so the twin moves at your rhythm rather than a stranger's. It is never shown as your face.
Angles. This decides how steady you look from the side.
Front, left, right, back, close-up. This step is optional and you can skip it, and you can add angles later from the twin's own page. Same outfit and same light across all of them, or it works against you.
Looks. This decides what you are wearing in the videos.
Four cards: keep my look, light tidy, groomed and styled looks. Light tidy is chosen for you, and it is the safe one. A twin gets three sets of starter stills, so you have the first set and two more goes.
Voice. This decides how it sounds.
Pick a voice from the library and you are finished. Or hand over a sample of your own and it speaks in yours, which takes ten seconds of clear audio. Full-length videos in your own twin start on the Plus plan. Cloning your own voice is on every plan, Free included.

Start with the photo

Then you finish, and the twin starts getting itself ready in the background while you go and write something.

The first run asks the same things in a different order

If you have just signed up, the first run walks you through it before the Studio opens, and it asks in this order: photo, look, name, recording, voice. The paid builder, the one you open later from your twins, asks photo, record, angles, looks and voice.

Same five things. Different order, because the first run puts the fun part early and the paperwork after it, and the builder puts consent immediately after the face it belongs to. Nothing has gone wrong if you recognise the steps but not the sequence.

What you actually say, and how long for

You do not write anything. The words are on the screen and they scroll, the way a teleprompter does.

Fifteen seconds is enough, and ninety is the most.
A take under fifteen comes back and asks for a few seconds more. A take over ninety asks you to keep it shorter. Nothing between those two numbers is better than anything else.
Nothing is charged for recording.
Recording, or re-recording, a consent clip costs no credits. You can go again as many times as you want, and the take you keep is the one on file.
Speak a little louder than usual, and a little slower.
The code is two digits in the middle of a sentence, and it is the part people rush.

If you would rather use your phone, the screen hands the recording off to it and picks the take up again when it is done.

What the face check is, and what a no means

Before the twin can be used, the face in the recording is compared with the face in the photo. It is one comparison between two images you gave us, and it is there for one reason: so that a photo of somebody else cannot be turned into a twin by somebody who is not them.

One encrypted face template is kept per person, and the readings taken for each check are not kept.

A no is not a fault. The check is set to be cautious, which means it would rather ask you to record again than let a stranger through. In practice a no usually means one of four things: the light was behind you, the photo was a three-quarter angle and the recording was straight on, half your face was out of frame, or you were wearing something in one and not the other. Record again in the same light as the photo and it passes.

If the code was not heard clearly, the twin is still built and the clip is still on file. You are asked to read the code again when you have a minute. It is an invitation, not a wall.

The list that saves you a second attempt

What not to hand over, and why each one comes back.

A photo with two people in it.
The door refuses more than one face, and a group shot where the other person is small and blurry still counts as two.
A photo where something covers part of your face.
Sunglasses, a cap brim or a hand across the jaw all read as a face the door cannot take a clean reading of.
A heavy filter, a beauty-mode phone shot, or a retouched photo.
The twin will hold that look in every video, including the ones you make in two years.
A photo at an angle when the recording will be straight on, or the other way round.
This is the most common reason a face check asks for a second take, and it is entirely avoidable.
A file that is not a JPEG, PNG or WebP, or one over 10 MB.
Camera raw files and the format most phones save in are not read, so convert first.
A voice sample recorded in a car, a cafe, or a room with a fan.
A cloned voice is built from what it is given. ten seconds of clear audio in a quiet room beats a minute of a busy one. The voice half is Voice cloning, where it is explained properly.

And what not to say on the recording

Anything instead of the script.
Paraphrasing is fine, skipping the code is not, and the code is the part most people forget.
The words on behalf of somebody else.
If the face in the frame is not yours, the consent is not yours to give, and a twin built that way is one we take down.
A person's name, a client's name, or anything confidential.
The clip is kept as the consent record, so treat it as a document rather than a rehearsal.
A hedge in front of the consent line.
Say it plainly. It is a record, and a hedged record is a worse record.

What each refusal means, in plain words

These are the doors' own words, and none of them costs you anything to hit.

What each refusal from the twin builder means, and what to do about it
What you see	What it means	What to do
We couldn't detect a clear, single face in that photo	What it meansTwo faces, no face, or a face too dark or too small to read	What to doOne person, front on, in even light. A phone selfie at arm's length is usually enough
Photo too large, max 10 MB	What it meansThe file, not the picture	What to doExport it at a smaller size, or take a fresh one in the guided capture
Only JPEG, PNG, or WebP photos accepted	What it meansThe format	What to doSave it again as a JPEG. A phone photo in its own format needs converting first
That recording is under 15 seconds, so give it another go and speak for a few seconds more.	What it meansThe take, not the words	What to doPress record again and keep going a little longer
We couldn't hear your safety code clearly. Record it again when you have a minute, so your twin stays verified.	What it meansThe two digits in the middle of the script	What to doNothing is lost and the twin is still built. Read the code again when you have a minute
This photo doesn't match the consented person for this Digital Twin	What it meansThe check would rather ask twice than let a stranger through	What to doRecord again in the same light and at the same angle as the photo

What happens once the twin is ready

The twin starts getting itself ready in the background. The first thing it makes is a set of stills in the look you picked, so you can see what it will look like before you spend anything on a video. The first run's sets of looks spend no credits. If you do not like the set, you have two more goes at it.

On Free, the first run ends on a four second sample. It is short on purpose: it is there so you can see and hear the thing before you decide anything. Full-length videos in your own twin start on the Plus plan. Cloning your own voice is on every plan, Free included.

After that the twin is just there, in your Workspace, and making a video is writing a script and picking it. AI talking head generator is where that happens.

Consent does not last forever.
A Digital Twin's consent lasts 365 days, and the twin stops rendering until it is given again. It is given again in the same fifteen seconds.
You can delete it.
Deleted Digital Twin data can stay recoverable for up to 30 days, and is then hard-deleted.
Nothing you hand over trains a model of ours.
Kyndrify does not use your photos, voice, scripts or finished media to train a Kyndrify model.

Building a twin for a principal, a partner or a client

Agencies and practice managers do this more often than individuals do, and the flow is built for it. You create the twin, and when the recording step comes you send a private link instead of recording anything. They open it on their own phone, read the same script, say the same code, and the take comes back to your Workspace.

Somebody who consents through a private link can withdraw it, and no new videos of them can be made after that.

The practical advice from people who do this often: book ten minutes rather than sending the link cold, tell them there are words on the screen so they do not write anything, and get the photo and the recording in the same room on the same day. A photo from one year and a recording from another is the pairing a face check is most likely to send back.

Consent, likeness and the part that is yours

A photo of your face, and the reading taken from it, are biometric information in some places, and the rules are not the same everywhere. What we do is keep a record: the recording, the date, and the notice that was in force when you gave it, which you can read in the biometric information privacy notice. A consent recording is kept for at least the life of the twin plus seven years, because it is the evidence the person agreed.

That is a record and not a certification. Two duties stay with you and not with us. If you publish a video of a person made this way, saying that it was made with AI is your call to make and your organisation's to answer for. And if the face in the twin belongs to somebody else, their consent has to be theirs, given by them, and they can withdraw it later.

Checked September 2026. This is a summary, not legal advice.

FAQ

Questions people ask before they start

Can AI create videos of yourself?: Yes, and this is how: you give a photo, a short recording where you consent out loud, and a voice. After that you type a script and the video comes back in your own face. There is no filming and no studio, and nothing is made that you did not write.
Do I need a video of myself to build a twin?: No, and you cannot use one. A twin is built from a still photo plus a short consent recording. If you have a video of yourself talking, it is not the input this takes.
How long does the consent recording have to be?: From fifteen to ninety seconds. Fifteen is genuinely enough. Anything under fifteen comes back and asks for a few more seconds, and anything over ninety asks you to keep it shorter.
What if the face check says no?: Record again in the same light and at the same angle as your photo. The check is set to be cautious on purpose, so it would rather ask twice than let a photo of somebody else through. It is not a mark against you, and recording, or re-recording, a consent clip costs no credits.
Can I build a twin of somebody else?: Only with their own recorded consent, given by them. You send them a private link, they record on their own device, and they can withdraw it later. You cannot record it for them, and a twin built from a photo of a person who has not consented is one we take down.
What happens to my photo and my recording?: A consent recording is kept for at least the life of the twin plus seven years, because it is the evidence the person agreed. One encrypted face template is kept per person, and the readings taken for each check are not kept. Kyndrify does not use your photos, voice, scripts or finished media to train a Kyndrify model. Deleted Digital Twin data can stay recoverable for up to 30 days, and is then hard-deleted.

Keep going

Ready to build yours?

One photo, fifteen seconds, and a voice. Free includes one Starter Twin for your Workspace, and 8,400 credits to spend on it.

Get started for free