数字人 explained: what an AI avatar is, how it’s made, and when it works

What a 数字人 actually is

A 数字人 (digital human, AI avatar, virtual presenter — the names overlap) is a video version of a real person, built from footage of that person, that can deliver new lines without them filming again. You type a script; the avatar says it, in your voice or a chosen voice, with your face and gestures. The output is an ordinary video file you post like any other.

Two things it is not. It is not a cartoon or a generic stock character — the useful kind is built from you, so customers see the person they already know. And it is not a chatbot — it doesn’t answer questions live; it reads what you wrote. Think of it as a reusable recording of yourself rather than a robot that thinks for you.

How it’s made: film once, script forever

The process has three parts, and only the first needs you in the room.

1. One filming session. You sit in front of a camera for a few minutes and talk — naturally, with your normal gestures, in the clothes and setting you want the avatar to live in. Good light and clean audio matter far more than an expensive camera; this footage is the avatar’s entire vocabulary of movement, so it’s worth doing properly. If you want a Chinese-language avatar and an English one, you record a short sample in each.

2. Building the avatar. The footage is processed into a model of how you look and move when you speak, plus a clone of your voice from the audio. This is a one-time step. You review the result and approve it — or re-film if the light was bad or you looked stiff.

3. Scripts, forever. From then on, every new clip is a script. Someone writes 60–90 words, the avatar delivers it, an editor adds captions, your logo and a cover frame, and it goes out. A new announcement takes minutes of your attention instead of a morning.

What it costs you in time

This is the actual reason SMEs use it. Filming a proper 口播 video costs a business owner half a morning: hair, setup, retakes, and the mental load of performing. Most owners manage it once, post three videos, and stop. The feed goes quiet, and a quiet feed tells customers the shop might be quiet too.

With an avatar, the owner’s time drops to: one filming session up front, then a few minutes a week approving scripts. The posting rhythm stops depending on whether the owner had a good hair day. That consistency — showing up every week in the brand’s own colours and voice — is what builds recognition, and it’s the thing most small brands can’t sustain with real filming alone.

Where it works well

• Announcements. New opening hours, a public holiday closure, a new branch, a price change. Short, factual, needs a face to feel personal.
• Promotions and reminders. “This week only,” “last few slots,” a seasonal push. The kind of post you should make every week and never do.
• FAQ replies. The five questions customers ask every day, each answered in 30 seconds. Post once, pin, and stop typing the same answer in DMs.
• Product explainers. What’s in it, how to use it, who it’s for. The avatar presents; product shots carry the visual.
• Two languages without two filming days. The same script in English and 中文, from the same face, for the Singapore audience that switches between them.

The common thread: short, clear, low-emotion content where the point is the information and the face is there to make it feel like it came from a person.

Where it looks fake — and why

People can usually tell within a few seconds when a clip is an avatar. That’s not always a problem — more on that below — but it becomes one in these situations:

• Emotion. A thank-you to a long-time customer, a story about why you started the business, an apology. The avatar’s expressions are borrowed from a ten-minute session; they don’t match a moment that needs real feeling.
• Long monologues. Past about 90 seconds the sameness of gesture and rhythm starts to show. Keep clips short.
• Reacting to things. Unboxing, tasting, trying on, demonstrating with hands. The avatar can’t hold your product.
• Generic scripts. This is the biggest one. An avatar reading marketing-speak that could belong to any shop looks like exactly what it is. An avatar saying something only your business would say — in the words you’d actually use — reads as you on a busy day.

The real you is still the strongest post

We’ll say this on every page where the avatar comes up: the most persuasive thing you can post is still you, on camera, in your own shop, talking to a customer. Avatars exist to fill the weeks you can’t do that, not to replace the weeks you can. A sensible month looks like one morning of real filming that produces four to six real clips, with avatar clips carrying the announcements and reminders in between. Customers get the real person for the moments that matter and a familiar face for the rest.

The businesses that get it wrong go all-avatar, post every day, and wonder why engagement falls. Frequency without presence is noise.

How to start without embarrassing yourself

Three rules. First, film the base session like it matters, because it does — a tidy background in your brand’s colours, daylight or a soft light, a lapel mic, and five minutes of you talking normally about what you actually sell. Second, write scripts in your own voice, in the language your customers use, and keep them under 90 words. If your assistant or an AI drafts them, you still read every one before it goes out. Third, start with announcements and FAQs, watch what customers say for a month, and only then decide how much of your feed the avatar should carry.

If you’d rather not run any of this yourself, it’s part of our branding & social media service: we plan the month, script it, film the real you once, and the avatar covers the gaps — in your brand’s look, not ours. Our smart tools page explains where we draw the line between what a tool does and what a person still signs off.

Common questions

What is a 数字人 / digital human?

A digital human, or AI avatar, is a video version of a real person made from footage filmed once. It can deliver new scripts in that person's face and voice without them filming again. The output is a normal video file you post to social media like any other.

Does an AI avatar need me to film every time?

No. You film one base session of a few minutes. After that, every new clip is just a written script the avatar delivers. Your ongoing time is approving scripts, not sitting in front of a camera.

Will customers know it's an avatar?

Often, yes — especially in longer clips. That's fine for announcements, promotions and FAQ answers, where the information is the point. It's not fine for emotional or story-driven posts, which should always be the real you on camera.

Can the same avatar speak English and Chinese?

Yes. With a short sample recorded in each language, the same face can deliver the same script in English and 中文, which suits a Singapore audience that moves between both.

Is a 数字人 a replacement for filming myself?

No. It fills the weeks you can't film. The strongest post you have is still you, in your shop, on camera. A good month is one real filming session plus avatar clips for the routine announcements in between.

Related: our branding & social media marketing service and the smart tools we use behind it.