An AI avatar video shows a digital copy of you, built from one recording of your face and voice, delivering scripts that you wrote or approved. For a founder who dreads filming, it can take over the routine talking-head work: the weekly LinkedIn post, the course lesson, the product explainer. It cannot replace you where the whole point is that you showed up in person, such as a live Q&A, an apology or the story of why you started the company.
What an AI avatar of yourself can and cannot do
| An avatar can | An avatar cannot |
|---|---|
| Deliver a script to camera with your face and voice | Decide what is worth saying |
| Record a new video while you are on a plane | Improvise, take questions or react to the room |
| Re-record one sentence when a fact changes | Carry real emotion in a high-stakes moment |
| Speak a language you do not, with matching lip movement | Walk through your real office or handle your real product |
| Keep a posting schedule through your busiest month | Be the person people meet on the sales call afterwards |
The last row matters more than it looks. A personal brand is a promise that the person in the video is the person who turns up. An avatar that is smoother, calmer and better lit than you are sets up a small disappointment at every first meeting. Aim for an avatar that looks like you on a good day. Nothing more.
What you need to record once
A consent statement. Yes, even for your own face. Avatar tools ask for proof that the person on screen agreed. HeyGen, for example, says that every video-based digital twin needs a short consent video recorded by the person shown. If a studio builds the avatar, you should also sign a usage agreement that says the studio may use your likeness for your videos and for nothing else. The same agreement should say who keeps the avatar’s files, and our explainer on who owns AI-generated video lists what to get in writing.
Footage to camera. Requirements vary by tool. HeyGen’s guide asks for a video of about two minutes and recommends natural movement without exaggerated gestures. When we build an AI presenter or avatar at Zero Shot, an AI video studio in London, we record more than the minimum: two or three outfits, seated and standing, with pauses where your mouth is closed. Good light, a plain lens at eye level, eyes on the camera and clean audio matter more than an expensive camera.
Voice samples. Covered in the next section.
Examples of how you really talk. Send a few unpolished voice notes or an old podcast appearance. The script has to sound like you, including the phrases you overuse.
After that one session you should not need to be filmed again until your appearance changes noticeably. A new haircut is fine. A new beard is a new recording.
How voice cloning fits in, and the consent and security it needs
The voice is what makes an avatar convincing to people who know you, so it deserves as much care as the face. There are three options.
- An instant clone. ElevenLabs recommends one to two minutes of good audio for its instant voice cloning. It is quick and good enough for short social clips.
- A professional clone. ElevenLabs asks for at least 30 minutes of clean audio and says that two to three hours is closer to ideal. Its documentation also states that you can only create a professional voice clone of your own voice, and it makes you pass a voice verification step before training starts.
- No clone at all. Record the audio for each script yourself and let the avatar lip-sync to your real voice. Many founders who hate the camera are perfectly happy with a microphone. This gives the most natural result and removes most of the security questions.
If you do clone your voice, treat the model like a password. A convincing copy of a founder’s voice is exactly what a fraudster needs to phone your finance team and ask for an urgent payment. Sensible precautions:
- keep the voice in an account that you control, with two-factor authentication
- limit who can generate audio with it, and check the usage history
- put it in the contract that a supplier may not use the voice for anything else, and must delete it on request
- tell your team and your family that a copy of your voice exists, and agree a call-back rule for any request involving money
- if your bank uses voice recognition, ask what other options it offers
A recording of your face and voice is your personal data, so you are entitled to ask where it is stored and for how long. This is general information and not legal advice.
How to stop an AI avatar video looking uncanny
- Keep takes short. Avatars hold up well for a sentence or two. Cut often, as you would with real footage.
- Use a medium frame. Head and shoulders, not an extreme close-up. Teeth and eyes are where generated faces give themselves away.
- Write the way you speak. Contractions, half-finished thoughts, your usual opening line. A flawless script read by a copy of your face is the fastest route to the uncanny valley.
- Cover half the video. Screen recordings, product shots, captions, charts. The avatar introduces and links, and the cutaways do the explaining.
- Do not ask for what it cannot do. Big gestures, walking and talking, picking things up. Stay close to what was recorded.
- Listen before you look. A flat, evenly paced voice gives the game away sooner than the picture does. Regenerate lines until the stress falls where you would put it.
- Throw away bad takes. A glitch in one frame is enough for the comments section.
- Mix in the real you. Real photos, a real clip from an event, a filmed video once in a while. The mixture keeps the avatar anchored to a living person.
What should you tell your audience?
Start with the rules, because they are the minimum. Meta has said it requires people to label organic posts that contain photorealistic video or realistic-sounding audio that was digitally created or altered. TikTok requires labels on AI-generated content that contains realistic images, audio or video. YouTube requires creators to disclose content that makes a real person appear to say or do something they did not, which is a fair description of an avatar of you. YouTube’s list of exceptions does include cloning your own voice for voice-overs, so a voice-only clone is treated more lightly there. If the video is an advert in the UK, the ASA’s position is that there is no blanket duty to disclose AI, but the ad must not mislead. We go through each platform, and the separate rules for ads, in do you have to label AI video.
Then go further than the rules. A personal brand runs on trust, and our advice is to say it once, plainly, somewhere permanent such as a pinned post or your bio: some of my videos use an AI avatar of me, and I write or approve every word. In our view, audiences forgive a tool far more readily than a secret. Platform policies change, so check the current wording before you rely on this paragraph.
When is filming yourself still the better choice?
- Launches and funding news. People want to see that you are excited, and they can tell.
- Apologies and bad news. Never send a copy to say sorry.
- The founder story. Why you started, what went wrong, what you learned. This is the content a personal brand is built on, and it has to be real.
- Customer conversations and events. Real people and real places belong on camera. Our guide to AI video or a real shoot covers this in more detail.
- The first months of a new personal brand. Let people meet the real you before they meet the copy.
There is a middle path that suits most camera-shy founders. Film one good session a quarter for the videos that need you, and let the avatar handle the routine in between. If you would prefer not to appear at all, there are two other routes: reels built from generated scenes with your real voice over the top, or an invented presenter who speaks for the company, which we describe in our guide to AI spokesperson videos.
If you would like an avatar of yourself, or want to talk through whether you need one, send us a short brief by WhatsApp or email. Tell us where the videos will run and how often, and we will reply with a fixed price.
Questions people ask
How much footage do I need to make an AI avatar of myself?
It depends on the tool. HeyGen's own guide asks for a video of about two minutes, plus a short consent video recorded by the person shown. A studio will usually record more than the minimum, in more than one outfit and setting, so the avatar has some range.
Will people be able to tell that it is an avatar?
Often, yes, especially people who know you. Short takes, a medium frame, a script written the way you speak and plenty of cutaways make it far less noticeable, but you should plan on telling your audience anyway.
Is it safe to clone my voice?
It is safe if you treat the voice model like a password. Keep it in an account with two-factor authentication, limit who can use it, make sure any supplier is contractually barred from other uses, and warn your finance team that a convincing copy of your voice exists.
Do I have to label AI avatar videos on Instagram, TikTok and YouTube?
In most cases, yes. Meta requires a label on organic posts with photorealistic video or realistic-sounding audio that was digitally created, TikTok requires labels on realistic AI-generated content, and YouTube requires disclosure when a real person appears to say something they did not.