Your own AI influencer in three steps

AI influencers already exist. They sell, they pull millions of views, and most people never notice that the face on screen does not exist.

But this guide is not about fooling anyone. The goal is a character so realistic and so consistent that people keep watching, comment, and buy. This is how creators are already scaling content without a weekly shoot.

Here is the whole process: design the face, shoot your base video, swap yourself out, put it to work.

Making a random pretty face is easy. The hard part is a face that still looks the same three months from now, and that your audience recognises in the first second.

A face that works for months has three parts.

• One memorable but not exotic feature. A small mole, a scar through an eyebrow, slight asymmetry. Something people remember without thinking about it. Not blue hair and violet eyes.
• Consistent lighting. Every reference image uses the same soft, even light. The model recognises the face through the light.
• A neutral expression as the base. Smiles and emotion come later, from the video. The base is a calm face, so every edit starts from the same point.

Think of a passport photo, just alive. That is the foundation, not a portfolio shot.

In Higgsfield, open the Soul model: every prompt in this part is written for Higgsfield Soul. If you do not have Soul, the same prompts also work in ChatGPT, the skin just comes out a little smoother. Ratio: 3:4. Seedance only enters in part three, when you edit the video.

01Prompt anatomy

Every face prompt has the same skeleton. Fill in the square brackets and paste it into the Higgsfield Soul prompt field.

A single head and shoulders photograph of a [AGE] year old [REGION OR ETHNICITY] [woman / man], front facing, looking straight into the lens.
Overall impression: [one line: who is this person, what do people trust them with].
Signature feature: [ONE memorable but not exotic feature, e.g. a small flat mole above the left corner of the mouth].
Hair: [length, colour, texture, how it sits].
Eyes: [colour]. Brows: [shape, thickness].
Skin: real skin texture with visible pores, slight natural asymmetry, faint uneven tone, no retouching, no makeup look.
Expression: calm and neutral, mouth closed, relaxed jaw.
Framing: shoulders up, camera at eye level, 85mm lens, f/2.8.
Background: plain neutral light grey seamless background.
Light: soft even front light, no hard shadows, no colour cast.
One person alone in frame, photorealistic, no text, no watermark.

02Six ready made characters

Three women, three men, all written for Higgsfield Soul. Four are built for the Armenian market, two, Maya and Daniel, for English speaking TikTok and Instagram, in case your product sells beyond the border. Take yours, rename them if you like, and generate. The first time, make 4 variations and pick the one you would follow yourself.

Lilit, 29

Calm, trustworthy skincare face. Sells cosmetics, skincare, wellness, anything where trust builds slowly.

A single head and shoulders photograph of a 29 year old Armenian woman, front facing, looking straight into the lens.
Overall impression: calm, rested, the kind of face people trust with skin advice.
Signature feature: a small flat mole just above the left corner of her mouth.
Hair: dark brown, straight, shoulder length, tucked behind one ear, a few loose strands.
Eyes: dark hazel. Brows: full, natural, soft arch, slightly uneven in thickness.
Skin: olive undertone, real skin texture with visible pores on the nose and cheeks, faint natural shadow under the eyes, slightly uneven tone on the forehead, no makeup look, no retouching.
Expression: calm and neutral, mouth closed, relaxed jaw.
Framing: shoulders up, camera at eye level, 85mm lens, f/2.8.
Background: plain neutral light grey seamless background.
Light: soft even front light, no hard shadows, no colour cast.
One person alone in frame, photorealistic, no text, no watermark.

Maya, 24

Energy, street style, fast talker, in English. Built for international TikTok: sneakers, clothing, accessories, anything bought on impulse, in any country.

A single head and shoulders photograph of a 24 year old Brazilian woman, front facing, looking straight into the lens.
Overall impression: quick, playful, street smart, looks like she is about to say something funny.
Signature feature: a thin faded scar cutting through the outer third of her right eyebrow.
Hair: dark brown, long, thick and wavy, pulled over one shoulder, a little untidy.
Eyes: dark brown. Brows: thick, straight, the right one interrupted by the scar.
Skin: warm brown tone, light freckles across the nose and upper cheeks, visible pores, a small healing blemish on the chin, no retouching.
Expression: calm and neutral, mouth closed, relaxed jaw.
Framing: shoulders up, camera at eye level, 85mm lens, f/2.8.
Background: plain neutral light grey seamless background.
Light: soft even front light, no hard shadows, no colour cast.
One person alone in frame, photorealistic, no text, no watermark.

Anahit, 38

The older sister you ask when you do not know what to do. Sells home goods, kids' products, education, services, money and health topics.

A single head and shoulders photograph of a 38 year old Armenian woman, front facing, looking straight into the lens.
Overall impression: warm, grounded, the older sister everyone asks for advice.
Signature feature: a small raised mole on the right cheekbone, and fine natural lines at the outer corners of both eyes.
Hair: dark brown with a few visible grey strands at the temples, pulled back into a low loose bun.
Eyes: grey green. Brows: medium, softly arched, natural.
Skin: light olive tone, real skin texture with visible pores, faint lines on the forehead, slightly darker skin under the eyes, no retouching, minimal natural makeup.
Expression: calm and neutral, mouth closed, relaxed jaw.
Framing: shoulders up, camera at eye level, 85mm lens, f/2.8.
Background: plain neutral light grey seamless background.
Light: soft even front light, no hard shadows, no colour cast.
One person alone in frame, photorealistic, no text, no watermark.

Daniel, 31

Clean, precise, no wasted words, in English. Built for an international audience: tech, apps, SaaS, software services, anything that sells across borders and needs explaining.

A single head and shoulders photograph of a 31 year old Northern European man, front facing, looking straight into the lens.
Overall impression: precise, clean cut, explains things without wasting words.
Signature feature: asymmetric eyebrows, the left brow sits visibly higher than the right.
Hair: light brown, short, neat on the sides, slightly longer on top, combed back.
Eyes: blue grey. Brows: light brown, medium thickness, the left one higher. Light stubble, two days of growth, evenly trimmed.
Skin: fair tone, real skin texture with visible pores, faint redness on the cheeks and along the jaw, light freckles on the nose, a small mole on the left side of the neck, no retouching.
Expression: calm and neutral, mouth closed, relaxed jaw.
Framing: shoulders up, camera at eye level, 85mm lens, f/2.8.
Background: plain neutral light grey seamless background.
Light: soft even front light, no hard shadows, no colour cast.
One person alone in frame, photorealistic, no text, no watermark.

Davit, 44

Warm café owner who remembers everyone's name. Sells food, drinks, restaurants, local produce, handmade goods, anything where the story sells.

A single head and shoulders photograph of a 44 year old Armenian man, front facing, looking straight into the lens.
Overall impression: warm, unhurried, a café owner who remembers every regular by name.
Signature feature: deep smile lines on both sides of the mouth and a short salt and pepper beard, more grey on the chin than on the cheeks.
Hair: dark brown going grey at the temples, short, slightly wavy, pushed back.
Eyes: warm brown. Brows: thick, dark, a little unruly, fine crow's feet at the outer corners.
Skin: medium warm tone, real skin texture with visible pores, faint broken capillaries on the nose, sun lines on the forehead, no retouching.
Expression: calm and neutral, mouth closed, relaxed jaw.
Framing: shoulders up, camera at eye level, 85mm lens, f/2.8.
Background: plain neutral light grey seamless background.
Light: soft even front light, no hard shadows, no colour cast.
One person alone in frame, photorealistic, no text, no watermark.

Tigran, 27

Gym guy, direct, motivating, no ego. Sells sports nutrition, activewear, training programmes, healthy products.

A single head and shoulders photograph of a 27 year old Armenian man, front facing, looking straight into the lens.
Overall impression: direct, energetic, disciplined, the gym friend who actually shows up.
Signature feature: a nose that is very slightly crooked, as if broken once, and a small mole below the left ear.
Hair: dark, very short buzz cut, sharp natural hairline.
Eyes: dark brown. Brows: thick, low, straight. Broad neck and shoulders visible in frame.
Skin: medium tone, slightly flushed cheeks, real skin texture with visible pores, a faint old scar on the chin, no retouching.
Expression: calm and neutral, mouth closed, relaxed jaw.
Framing: shoulders up, camera at eye level, 85mm lens, f/2.8.
Background: plain neutral light grey seamless background.
Light: soft even front light, no hard shadows, no colour cast.
One person alone in frame, photorealistic, no text, no watermark.

03The consistency step: the reference set

With the face you chose, make 5 more images: three quarter left, three quarter right, profile, a small smile, mouth open (mid word). With the first image, that is 6.

Every image is generated with the first image as the reference, never from scratch. Otherwise you get a new person each time. Upload the first image as @image1 and use this prompt, changing only the [ANGLE] line.

@image1 is the character. Face, hair, skin marks and identity exactly per ref. Do not change the background or the lighting.
A single head and shoulders photograph of the same person as @image1, [ANGLE].
Same person, same age, same hair, same signature feature. Mouth closed unless the angle line says otherwise.
Same soft even front light, same plain light grey seamless background, camera at eye level, 85mm lens, f/2.8.
Real skin texture with visible pores, no retouching. One person alone in frame, photorealistic, no text, no watermark.

The five [ANGLE] options:

• Three quarter left: rotated 45 degrees to their left so the front and the left side of the face are both visible, head turned to face the lens
• Three quarter right: rotated 45 degrees to their right so the front and the right side of the face are both visible, head turned to face the lens
• Profile: turned 90 degrees so the left side of the face is in a direct profile, looking straight ahead and not at the lens
• Small smile: front facing, looking into the lens, a small natural closed mouth smile, eyes slightly narrowed
• Mid word: front facing, looking into the lens, mouth slightly open as if in the middle of saying a word, relaxed jaw

Keep the 6 images in one folder with plain names: lilit_01_front, lilit_02_left and so on. In Higgsfield, upload the same images as reference elements so they are one click away every time.

This is where people stop. "I wanted to stay off camera, and now you are telling me to film myself." Yes. But nobody will ever see you.

The AI needs real human movement: pauses, hand gestures, speech rhythm, the way your head tilts slightly when you stress a word. No model can generate that from nothing yet. You are the skeleton, the character is the skin. The better the skeleton moves, the more real the skin looks.

Shooting checklist:

• Phone, vertical 9:16, highest quality setting.
• Camera at eye level. Not from below, not from above.
• Even light from the front: a window or a ring light. No shadow from the side.
• Plain, single colour background. No objects, nothing moving.
• Face fully visible, shoulders up, centred in frame.
• Hands never cross the face. Gestures stay at chest height.
• No sunglasses, no regular glasses, no hat.
• No hair over the face. Tie it back if it is long.
• 10 to 20 seconds per video. No longer.
• Speak the actual script, not "I will add the voice later". The lip rhythm comes from here.

Your voice can stay or be replaced later. This guide covers the visual part only.

Now the two halves meet: your video and the character's 6 images. Now it is time for Seedance: in Higgsfield, open Seedance 2.5 and switch to video edit mode.

Edit video. @video1 is the base video. @image1 to @image6 are the same person: the character. Face, hair, skin marks and identity exactly per ref. Do not use the image backgrounds or lighting.
Replace the person in @video1 with the character from @image1 to @image6.
Keep the exact motion, timing, head movement, lip movement, eye direction, framing, camera position and lighting of @video1.
Do not change the background, the hands, the clothing or the performance rhythm. Only the face and head change.
Skin texture on the character must match the references: visible pores, no smoothing, no retouching.
Keep the original audio unchanged. No subtitles.

Review checklist before you post:

• The face is the same from the first second to the last. No frame where "someone else" shows up.
• The lips match the audio, especially at the start of words.
• The hands look human: five fingers, natural movement, nothing "floating".
• Skin texture matches the reference set. No sudden shine or smoothing.
• The mole, the scar, the brow, your signature feature, stays in place for the whole video.
• Background and clothing have not changed.

When it fails. The first reflex is to rewrite the prompt. Don't. Nine times out of ten the problem is the base video. Face falls apart on a turn: reshoot with slower movement. Skin looks flat: the light was too low. Lips out of sync: you spoke too fast or part of your face was covered. Reshoot in 30 seconds and try again. Change the prompt only when the same video gives the same error twice.

Three formats that work right now, in Armenia and beyond it. The big advantage of a character is exactly this: it has no border. Same product, same video, voice in English, and your TikTok is suddenly in a market of hundreds of millions instead of three.

• Product in the character's hands. In the base video you hold your product, show it, open it. After the edit the character holds it. Hands and product stay yours, which is exactly the point: not one pixel of the product changes.
• Talking head tips series. One topic, one tip, 15 seconds. For a cosmetologist, an accountant, a coach, a furniture maker. The character says five times a week what you explain to customers every day.
• "One thing I learned today". A short daily format where the character shares one observation from your field. Easiest to shoot, builds the most connection, because the audience starts to know the character as a person. Two versions of the same video, Armenian and English, and one shoot covers two audiences.