HeyGen Digital Twin: How to Make an Avatar of Yourself

A Digital Twin is HeyGen's video-trained avatar of a real person. You film yourself talking, HeyGen trains on that footage, and afterwards you can type a script and get a video of yourself saying it. It is the feature most people actually signed up for.

Most failed attempts fail at the filming stage, not the software stage. The training footage requirements are specific, and a recording that breaks one of them produces a twin that looks slightly wrong in a way you cannot fix afterwards.

So read the whole filming section before you press record. Twenty minutes of preparation saves you a rebuild, and rebuilds are rate limited, which we will get to.

Whose face are you allowed to use?

Your own, or someone else's only with their explicit permission. This is the part of the process that is not a preference, and it is worth being clear about before anything technical.

HeyGen's moderation policy states that creating avatars of other individuals is strictly prohibited without permission, that consent must be obtained from the person depicted, and that removal requests must be honoured at any time. It also prohibits avatars of real individuals including celebrities and public figures without their explicit consent, regardless of how famous they are.

Their ethics page puts the same position in one line: avatars and the videos depicting them should only be created or shared with expressed permission. That is a good rule to hold yourself to independently of any platform enforcing it, because the legal exposure of a synthetic video of someone who did not agree sits with you, not the tool.

What footage does HeyGen need?

One continuous, unedited recording of at least two minutes, filmed at 1080p or better, with you speaking to camera. HeyGen's Digital Twin article recommends up to five minutes for stronger training, at least 1080p and 30fps with 4K at 60fps preferred, and supports landscape 16:9 or portrait 9:16 orientation.

The audio matters even if you plan to use a different voice later. Their documentation notes the footage must include you speaking so the system has something to learn lip movement from. Silent footage is not usable training material.

How should you perform on camera?

Still, calm and looking straight down the lens. HeyGen's filming tips are unusually prescriptive here, and the prescriptions are all about giving the model a clean, consistent subject rather than an expressive one.

• Hold eye contact throughout. Their guidance says to maintain direct eye contact with the camera at all times and avoid looking around or up and down.
• Keep your hands below chest level. Gestures are fine within frame, but raising hands above the chest is called out specifically as something to avoid.
• Start with a pause. Begin the recording with a 2 to 3 second pause, and insert 1 to 2 second pauses between longer sentences, with your lips closed and eye contact held.
• Stay still. Avoid sudden head motions and unnatural body movement. Excessive movement in the source footage shows up in how the twin behaves later.
• Do not memorise. Their article says to speak naturally and that memorisation is not required. A teleprompter placed right beside the lens works better than a recital.

Film the consent video in the same session, in the same room, in the same clothes. HeyGen's documentation states the consent video must be recorded and submitted at the same time as the avatar footage, so treating it as a separate errand for next week just stalls the upload.

What exactly is the consent video?

A recording under 30 seconds of the same person reading HeyGen's provided consent text exactly as written. Their help centre is direct that no paraphrasing is permitted, and that videos which are too dark, blurry or have unclear audio get rejected.

Creating the Digital Twin, end to end

1. Set up your filming space — Quiet room, bright even lighting, plain static background. Camera at eye height. Clothing without large logos or text. Confirm your camera is set to at least 1080p and 30fps before you record anything.
2. Record the training footage in one take — At least two minutes, up to five for better training, in a single continuous shot with no cuts. Speak naturally to the lens, hold eye contact, keep gestures below chest level, and pause with lips closed between longer sentences.
3. Record the consent video in the same session — Under 30 seconds, reading HeyGen's provided consent text word for word. It must be the same person as in the footage. Accepted formats are MP4, MOV or WebM, from 480p up to 4K, up to 10GB.
4. Open the Digital Twin flow — In HeyGen, go to Avatars, then Create New Avatar, then Digital Twin. From there you record or upload your video.
5. Upload footage and consent together — HeyGen documents three consent submission routes: upload a pre-recorded file or paste a Google Drive link, record live with a webcam, or scan a QR code to record on a phone. The QR code route is also how you send the consent request to a different person.
6. Wait for training, then check the avatar card — HeyGen's article gives roughly 10 to 20 minutes of processing for non-4K footage. If something is wrong, rejection reasons appear directly on the avatar card in your Avatars page, which is the first place to look rather than support.
7. Test with a short script before you commit — Generate one short clip and watch the mouth on hard consonants and the transitions between sentences. If the twin looks off, the cause is almost always in the source footage, so note what to change before you use up your rebuild.

None of those steps is difficult on its own. It is a preparation checklist, and preparation checklists are the things that quietly do not get done, which is why so many accounts have an empty avatar slot.

That is the same complaint we set out to remove at Sistava, in a different domain. Instead of assembling a setup and then keeping it alive, you hire an AI Employee and describe the outcome you want in plain English. It has its tools connected already, it asks you when it is unsure, and it configures itself. Filming your own face is still on you.

How many avatars do you get, and can you redo one?

Slots are limited by plan and rebuilds are limited by time. HeyGen's Digital Twin FAQ lists one slot on the free plan, five on Creator and Pro, and ten on Business, with additional slots available as a paid add-on.

Redoing an avatar means deleting the current one from its slot and uploading fresh consent and footage, and their FAQ limits that to once per billing cycle. That single sentence is the reason to over-prepare the filming session rather than plan to iterate.

What do you feed the twin once it exists?

Scripts, constantly. A twin is a rendering engine for words, and the bottleneck moves immediately from filming to writing. People who build one and then post nothing almost always ran out of things to say, not out of credits.

That writing job is what a Sistava AI Employee does well next to a tool like this. You brief it once on your voice, your audience and your offer, and it drafts the week's spoken-word scripts, cut to the length you asked for, ready to paste in. It has your context already, and it asks you when it is unsure rather than guessing.

The same applies on the other side of publishing. Comments arrive, someone asks a question in a reply, a prospect watches and does nothing. An AI Employee can pick those up, answer them and follow up, which is where a video actually turns into a conversation.

Sources

• HeyGen Help Center, Create your first Digital Twin (Video Avatar)
• HeyGen Help Center, Digital Twin (Video Avatar): Filming Tips
• HeyGen Help Center, Recording your Consent Video
• HeyGen Help Center, Digital Twin FAQ
• HeyGen, Moderation Policy
• HeyGen, Data Handling, Ethics and Content Authenticity