AI Influencer Prompts: Copy and Paste, Keep the Same Face

The prompt that decides whether your AI influencer looks like the same person in every picture is the first one, the base character prompt, and almost none of it belongs in the prompts that come after it. This page gives you the prompts to copy and paste at each stage: the base character, three looks that each change one thing, and a first short clip, with what those looks and that clip cost at named settings before you press Generate.

The method fits in two sentences. Write the character once, in detail, and let the saved character carry the face from then on; every later prompt describes the scene and changes one thing per picture, so when something drifts you know which line moved it. If you are new to the idea, start with what an AI influencer is, and if you want the character maker walked end to end before you write a word, read the AI character generator guide first.

Key takeaways

• Write the face, hair, build and marks once, in the base character prompt, and leave them out of every scene prompt.
• Each scene prompt names the outfit, the place, the pose, the light and the framing, and changes one of them per picture.
• Influverse states that one character stays consistent across every model with no training step, from the first portrait to a 5 second Kling 2.1 clip. So write the face once, keep the Identity pill on, and leave the face out of every later prompt.
• Three GPT Image 2 images at Balanced quality cost 3 credits each, 9 credits for the set, and one 5 second Kling 2.1 clip costs 10 credits, so the first set of looks and one clip comes to 19 credits, $0.95 at $0.05 a credit.
• Wan 3.0 Prime takes the character's reference photos into a clip, up to 10 of them; Kling 2.1 takes 0 reference photos, so the only picture of your character it sees is the start frame you pick.

Before you begin

• An Influverse account. Sign up free and get 20 credits, no card needed.
• A niche and a name, so the base prompt has a style to describe. If you have the niche and no name yet, start with the AI influencer name generator.
• One idea for a first clip: a single action the character could do in a few seconds, such as walking toward the camera or standing up from a cafe step.
• The credit cost of the first set, which step 5 prices at a named model and quality, read against the credits you have.

What makes an AI influencer prompt work

A prompt has two halves, and the skill is keeping them apart. The character half is everything that must never change: age and build, face and skin, hair, the marks and jewellery that are always there, and the default style. The scene half is everything that changes from picture to picture: outfit, place and time of day, pose and action, light and lens, framing and aspect ratio.

Mixing the two halves in one paragraph is fine for a single picture and a problem for the fifth. Writing the face into every scene prompt gives each rewording a chance to move it, with the words and the reference picture competing to say what the face is. Keep the words for the scene and let the reference carry the face.

Write the character half into the description box when you create the character. Keep the Identity pill on when you generate, and write only the scene half into each later prompt. The four prompts below are built that way: none of them mentions the face, the hair or the marks after the first.

Step 1: Write the base character prompt

By the end of this step you have one paragraph that describes your character from head to toe and a first portrait made from it. Paste it into the description box when you create the character (start from the AI influencer creator if you do not have an account yet), then keep the paragraph: it is the character half of every prompt you will ever write for them.

Copy this and change the bracketed parts:

A [age]-year-old [woman or man] with a [build] build and a [tall, average or compact] presence. [Face shape] face, [skin tone] skin with [texture: freckles across the nose, a light tan, visible pores], [colour] [shape] eyes, [brow shape] brows, a [nose], [lips], a [jawline]. [Hair colour], [length], [texture], usually worn [styling]. Distinguishing marks that never change: [a small scar through the left eyebrow, a thin gold hoop in the right ear, a line tattoo on the inner left forearm]. Default style: [clean athletic basics in muted colours, one bold accessory]. Photographed as a plain studio portrait, soft daylight, neutral background, looking at the camera.

Filled in, for a fitness creator:

A twenty-six-year-old woman with a strong, athletic build and an upright presence. Oval face, warm olive skin with a scatter of freckles across the nose and visible pores, dark brown almond-shaped eyes, straight dark brows, a small straight nose, full lips with a slight asymmetric smile, a soft but defined jawline. Dark brown hair past the shoulders, thick and wavy, usually worn in a high ponytail. Distinguishing marks that never change: a thin scar through the outer left eyebrow, a small gold hoop in the right ear, a single line tattoo of a wave on the inner left forearm. Default style: clean athletic basics in muted greens and greys, one bold accessory. Photographed as a plain studio portrait, soft daylight, neutral background, looking at the camera.

Three things in that paragraph do the work. The marks, a scar, an earring and a tattoo with a side and a surface named, give every later picture something to match beyond a general impression. The default style is there for the scene prompts that say nothing about clothes. And the last sentence asks for a plain portrait rather than a scene, because this portrait is the picture you will keep as the reference, so keep it free of the hat, the sunglasses and the busy background you do not want in every later picture.

Verification: the portrait shows the marks you named, on the side you named. If it does not, fix the paragraph before you make anything else, because you will check every later picture against this one. If a model's safety check stops the description, reword the part it objects to or try another model the picker offers.

Step 2: Outfit, place and pose prompts that change one thing

By the end of this step you have three looks of the same character. Open the image page with your character selected and the Identity pill on, pick a model, and paste one scene prompt at a time. None of the three prompts below mentions the face, the hair or the marks.

Outfit. The first look changes only the clothes:

Full-length mirror selfie in a bright bedroom, phone held at chest height, slight head tilt. Wearing a cropped sage green zip jacket, black high-waisted leggings, white socks and chunky white trainers. Soft window light from the left, clean white walls, a plant in the corner. Casual phone photo, vertical portrait framing.

Place and time of day. The second keeps the outfit and moves the character:

Sitting on the steps outside a cafe at golden hour, iced coffee in one hand, sunglasses pushed up into the hair. Wearing the sage green zip jacket and black leggings. Warm low sun from the right, long shadows, a quiet street behind. Candid lifestyle photo, slight grain, vertical portrait framing.

Pose and action. The third changes what the character is doing and how it is lit:

Mid-stretch on a yoga mat in a sunlit studio, one arm reaching up, eyes closed, calm expression. Wearing a black sports bra and grey joggers. Hard side light, wooden floor, large windows out of focus behind. Shot from slightly below with a slightly wide lens, vertical portrait framing.

Each prompt changes one thing against the one before it: the outfit, then the place and light, then the pose. When a picture comes back wrong, you know which line to change. Change two things at once and you are guessing.

One scene image costs between 2 credits and 12 credits across the image models, priced per image by quality, and every generation shows its exact credit cost before you confirm. Each row below gives one image at the quality named.

All four models in the table above read reference photos: Nano Banana 2 reads up to 5, and GPT Image 2, Seedream 5.0 Pro and Seedream 5.0 Lite read up to 10 each. Keep the Identity pill on whichever one you pick.

The live table for every model and every quality, including the ones this page does not show, is on the models page. Pick one model for all three looks: a model change is one more thing that moved.

Step 3: Keep the same face across images

To keep the face put, do three things: use a plain portrait with the marks visible as the reference, keep the Identity pill on, and never re-describe the face in a scene prompt. GPT Image 2 at Balanced quality is 3 credits an image and Nano Banana 2 at 2K is 5 credits an image, so a model switch is a price decision. Keep the Identity pill on in both.

When the fifth picture still looks like someone else, check these first:

• Re-describing the face in a scene prompt. Every rewording is a small push away from the reference. Describe the scene and nothing else.
• A busy reference. Keep sunglasses, hats and heavy makeup out of the base portrait and put them in a scene prompt, so they are a choice rather than a fixture.
• Changing two things at once. A new outfit and a new place in one prompt leaves you unable to tell which one moved the result.
• Uploading reference photos that fight the identity. On Nano Banana 2 the model reads up to 5 reference images and on GPT Image 2 up to 10. Use the spare slots for an outfit or a product, never for a different face.

Step 4: Prompts for a first short clip

By the end of this step you have a first short clip of the character. Video prompts describe motion, and the sound on a model that makes it, never the face. Which picture of your character the model gets depends on the model. Wan 3.0 Prime, Seedance 2.5, Seedance 2.0, Seedance 2.0 Mini and Seedance 2.0 Fast generate from reference images, and Wan 3.0 Prime reads up to 10 of them; Kling 2.1 generates from one start image and takes 0 reference photos.

So the choice is practical. On Wan 3.0 Prime, keep the Identity pill on and write the clip as a scene. On Kling 2.1, pick the start frame yourself, a look from step 2 with the marks visible, and treat it as the picture of the character the clip will have.

For a reference model such as Wan 3.0 Prime, written as a scene:

She walks toward the camera along a quiet sunlit street, ponytail swinging, adjusts the strap of a gym bag on her shoulder and smiles as she passes. Handheld tracking shot, warm late afternoon light, soft street sound and footsteps, no music, no dialogue.

For a start-frame model (Kling 2.1), written as what happens next in the cafe-step frame from step 2:

She lowers the iced coffee, looks up from the steps toward the street, stands and turns to walk out of frame to the right. Slow push in, steady camera, natural light unchanged, no dialogue.

Write the motion as a short sequence a person could act in a few seconds, name the camera move once, and on a model that makes audio, say what the sound is. Leave out any line about the face, the hair or the outfit.

A 5 second clip costs 10 credits on Kling 2.1, which makes silent video, 13 credits on Wan 3.0 Prime at 480p and 42 credits on Seedance 2.5 at 480p, both with audio. Each video request is priced once from its model, length, resolution and audio setting, then rounded up to a whole credit, so check the figure the page shows for the length you pick rather than scaling the 5 second price.

Step 5: Price the first set before you generate

Here is the first set priced end to end, as a worked example with the settings fixed: the character already created, three scene images on GPT Image 2 at Balanced quality, and one short clip on Kling 2.1 started from one of those images.

Three GPT Image 2 images at Balanced quality at 3 credits each plus one 5 second Kling 2.1 clip at 10 credits: 19 credits, which at $0.05 a credit is $0.95.

The point where this stops working is the clip. Kling 2.1 takes 0 reference photos and starts from one image, so the start frame is the only picture of the character it gets. If none of your three looks is a clean, well lit picture of the character, make that frame first, because no clip prompt wording stands in for it.

Three GPT Image 2 images at Balanced quality and one 5 second Kling 2.1 clip are what the 19 credits above cover, and every generation shows its exact credit cost before you confirm. Before you create the character, read its own price on its page, since the worked example above starts after the character exists. Plan prices are on the pricing page.

You now have a base character prompt, three scene prompts that change one thing each, a clip prompt for each kind of video model, and the price of the first set at named settings. When the clip is good enough to post, the guide to making AI TikTok videos is where the posting routine starts.