Guide · 14 min read · Updated November 3, 2026
How to make an AI influencer, step by step
We made Mira, an AI character, on November 2, 2026, and took her from a text description to a 50-second video with dialogue. This guide is that workflow: the steps that worked, the ones that failed, and what each step costs on Klakix. Mira isn't a real person. Every image and clip of her on this page was generated, and we label her as AI wherever she appears.
The short version
- Write a fixed identity. Describe one fictional adult with two or three distinctive but natural marks, such as a beauty mark, freckles, an eye color and a signature hairstyle, and paste the exact same text into every prompt.
- Pick one anchor portrait. Generate a few candidate portraits, keep the one that fits, and use it as the reference for every image that follows.
- Build a character sheet from separate angles. Generate each angle as its own high-resolution image with the anchor as the reference, lay the panels out yourself, and drop any image where the identity mark moved.
- Test a new scene, then add outfits. Put the character in a new place and new clothes to check the face holds, and make outfit images from the same references.
- Turn the character into video. Use the portrait as a first frame, give the front, 3/4 and outfit images to a reference model, and keep one slow camera move per shot.
- Write long takes as a timeline. For a 15- or 30-second single take, list what happens in each window of the shot, one event per window.
- Give the character one voice. Pick one ready-made voice and keep it; for long dialogue, record the lines separately and sync them to the lips in your editor.
- Label the account and the posts. Turn on Instagram's AI-generated profile label and the AI label on realistic posts, and never present the character as a real customer.
What counts as an AI influencer
An AI influencer is a social media account built around a person who doesn't exist: the face, and often the voice, are generated. A real creator or company writes the posts and is responsible for them. Instagram draws the same line. Its AI-generated profile label is for accounts that feature an AI-generated person, not for creators who use AI tools to edit or caption their own content.
Step 1: Write a fixed identity
Consistency starts in the text. Describe one fictional adult once, pick two or three marks that are distinctive but natural, and paste the exact same description into every prompt. Mira's identity sheet runs to about 280 words; these lines did the most work:
“ONE small dark-brown beauty mark on her LEFT cheekbone, just below the outer corner of her left eye (it appears on the right side of the image when she faces the camera).”
“a light dusting of freckles across the nose bridge and upper cheeks”
“soft, loose S-shaped waves that are smooth and glossy (not curly, not frizzy, not voluminous)”
“small gold huggie hoop earrings in both earlobes; a very thin delicate gold chain necklace”
The beauty mark is the most useful line in the sheet. It sits on one side only, so it doubles as a check: if it shows up on the other cheek, the image was mirrored and you throw it away. Keep the person clearly fictional and adult: don't write that the character looks like someone, and don't start from a real person's photo.
Step 2: Generate candidates and pick one anchor portrait
We generated four candidate portraits from the identity sheet with GPT Image 2.5, which ranked first for both text to image (Elo 1222) and image editing (Elo 1199) on the Artificial Analysis leaderboards in November 2026. We kept the second one. That portrait is the anchor: from here on, every image of Mira is made with it as the reference.

1

2: kept

3

4
Shortcut: the character generator runs this step without a prompt. You pick the look from menus, it draws 4 candidates, and the one you save gets a 3/4 and a full-body angle automatically, for 35 credits with GPT Image 2.5 or 48 with Nano Banana 2.
Step 3: Build a character sheet from separate angles
A character sheet, also called a turnaround or reference sheet, shows the same person from every side. Our first try asked for the whole sheet in one image with eight panels. It looked fine at a glance, but in the small panels the face drifted: it got longer, and the smooth waves turned into curls. We threw it away and changed the method:
- Generate each angle as its own image at high resolution (1536 × 2048 for close-ups, 1536 × 2304 for full length) with the anchor portrait as the reference. For the full-length views we also added the front full-length image.
- Cover the angles you'll use: for close-ups, the front, 3/4 and profile from both sides, and the back; for full length, the front, 3/4, profile and back.
- Lay the sheet out yourself. We matched the gray background, the figure height and the floor line with a short script; any layout tool works.
- Check the identity mark on every panel, and drop mirrored images.

You don't need every angle for video. Our Seedance shots used the front portrait, the 3/4 close-up and one or two outfit images, which is close to what the character generator saves for you.
Step 4: Test the face in a new scene, then add outfits
Before you trust a sheet, put the character somewhere new in different clothes. Our test put Mira in a café over the Bosphorus in a knit cardigan. The freckles, the beauty mark, the eyes and the earrings all carried over, so the sheet was ready.
Outfits work the same way. For an Egypt sketch we made a costume sheet, two images of Mira in an ivory gown with gold jewelry, from the same identity references. In video, the costume goes in as its own reference image next to the face references.


Step 5: Turn the character into video
There are three ways to put the character in motion, depending on the shot:
- First frame: start an image-to-video clip from the portrait or a scene image. On Klakix, “Pick from my characters” puts your saved character in that slot.
- References: give a reference model several images (front, 3/4, outfit) and describe a new scene. Seedance 2.5 Reference takes up to 30 photos; you refer to them in the prompt as @Image1, @Image2 and so on.
- Talking to camera: for a presenter reading a script, Talking photo voices the text and lip-syncs the portrait.
Keep one slow camera move per shot. Our prompts said “smooth gimbal at walking pace, no fast moves, no whip pans”, and we never speed up a generated clip afterwards.
Seedance and realistic faces
Seedance follows ByteDance's policy and rejects photos of real people as references; some apps show this as a “visual restriction” error. Our AI-generated characters passed, but not every time. On November 2 it accepted Mira in 3 of 3 runs, as a first frame and as reference images, and a second AI character in 5 of 5. On November 3, for the medieval reel, it accepted 10 of 24 runs with Mira's face images; references without a face passed 6 of 6. Across both days, face images of an AI character passed in 18 of 32 runs. Also rejected:
- A multi-panel character sheet, once. Give the model the single angle images, not the composite.
- A product photo with a real model in it. We redrew that first frame with AI instead, so no real person was in it.
If a run is rejected, cut the face images down to one, give the outfit as crops that don't show the face, or send the same request again: the same request was rejected once and passed the second time. ByteDance can change its filter at any time. On Klakix, a generation the provider blocks counts as failed and the credits come back automatically. Model details are on the Seedance page.
Step 6: Write a long take as a timeline
Seedance 2.5 can generate 30 seconds in one take. For a 30-second shot of Mira touring a palace, we wrote the prompt as a timeline, and the model followed it beat for beat: tea served on a gold tray, a barber shaving the Sphinx, a courier asking for a signature, the throne, and a hand over the lens at the end. This is the structure, shortened, with the dialogue summarized (Mira and the cameraman spoke Turkish):
@Image1 and @Image2 show the same young woman, Mira: keep her face, freckles, the small beauty mark on her left cheekbone, hazel-green eyes and her own wavy dark-brown hair with curtain bangs exactly. @Image3 and @Image4 show her costume: keep it exactly. ONE CONTINUOUS SHOT, NO CUTS: a smooth gimbal camera follows Mira at walking pace through an opulent ancient Egyptian palace by the Nile at golden hour. Steady and calm camera, no fast moves, no whip pans. 0-5s: a sunlit courtyard with a long lotus pool. The off-screen cameraman asks if she's ready; she glances back and smiles. 5-10s: two servants bring a golden tray of tea in tulip-shaped glasses; Mira takes one without stopping. 10-17s: a hall of painted columns where a barber on a ladder shaves a giant Sphinx covered in shaving foam. 17-23s: in a treasure room, a courier in a plain grey uniform (no logos) asks her to sign a tablet. 23-30s: she sits on a golden throne, looks into the camera and covers the lens with her palm; the frame goes dark. Sounds: footsteps on stone, servants murmuring, tea glasses clinking, the razor scraping, distant birds.
Two lessons from this shot:
- Scene logic matters. A joke about how many stones were left fell flat in front of a finished pyramid. With a half-built pyramid, a ramp and Mira carrying tea to the workers, the same line made sense.
- Draft at 480p, then upscale. We generated at 480p and upscaled the keeper to 1080p with Topaz; it looked close to native 720p for about half the price. On Klakix, a 30-second Seedance 2.5 Reference take is 308 credits at 480p plus 38 for the 1080p upscale, against 692 credits at 720p. The upscale runs in the AI video upscaler.
The format came from a viral Kling 4.0 clip from October 31, 2026: a single-take tour full of absurd modern details, a running chat with the person behind the camera, and a closing line that asks viewers to comment a keyword to get the prompts. Our version swapped in local jokes for a Turkish audience.
Step 7: Give the character one voice
People recognize a character by voice as much as by face, so pick one voice and keep it. What we learned voicing Mira:
- A library voice beat a designed one. A voice we designed from a text description sounded over-acted. A professional ready-made voice sounded natural, and it became Mira's permanent voice.
- Short lines can come from the video model. Seedance said a short Turkish greeting word for word, but in the 30-second take with two speakers it garbled several words. We recorded the script again with ElevenLabs v4 and synced it to her lips.
- Keep the real laugh and the real room. The [laughs] tag sounded fake in this scene; the laugh Seedance generated matched her face, so we kept it, along with the video's own background sound.
On Klakix, the AI voiceover tool has ElevenLabs v4 and ready-made voices (1,000 characters cost 10 credits). There's no voice cloning, and Seedance Reference here doesn't accept speech as reference audio, so lining a voice up with a generated shot happens in your editor. Our full list of audio lessons, with the numbers, is in the voiceover guide.
Step 8: Label the account and the posts
Since August 31, 2026, Instagram's label for these accounts is called “AI-generated profile”; it replaced the earlier “AI creator” label. You turn it on from Edit profile. Instagram says accounts that add it keep their reach, while accounts that feature an AI-generated person without it can lose recommendations, the places where Instagram suggests content to people who don't follow you. Label realistic AI posts as well. Who needs the label and what happens without it: AI-generated profile on Instagram, explained.
Rules we follow with Mira
- Always labeled as AI. Mira never appears as a real customer, reviewer or patient. In the US, the FTC's rule on fake reviews and testimonials, in force since October 21, 2024, bans testimonials that misrepresent that the reviewer exists, AI-generated ones included.
- No real person's face. Mira was described from scratch. The character generator here has no photo upload, and the video tools that take a photo ask you to confirm you have the person's permission and that they aren't a celebrity.
- Adults only. The character is an adult and is written and dressed as one.
- Paid posts are disclosed with the platform's paid partnership label, the same as for a human creator.
- No made-up results. We don't publish follower or income numbers we can't verify, for Mira or anyone else.
A second video: a wordless medieval reel in 10 prompts
On November 3, 2026, we made a second video with Mira: no dialogue, 34 seconds, 10 shots. Each shot puts exactly one modern object in a 14th-century town: a supercar on the drawbridge, a cash machine in the castle wall, contactless payment at the blacksmith's, a drift in the market square, a money gun, champagne with a sparkler, and a dragon back from shopping. Every shot was generated with Seedance 2.5 Reference at 480p and 4:3 (752x560), and the edit was upscaled to 1440x1080 with Topaz. This is the section we send to people who comment “FLEX” on Instagram.
The rules behind the prompts:
- One modern object per shot. Clothes, places and extras are 14th-century, and nobody in the shot is surprised by the object; that is where the joke comes from.
- A reference line. The face, the outfit, the car, the dragon and the town come as separate images, and the first line of every prompt says what each @Image is and what must stay exactly the same.
- A short timeline. Shots of 4 to 8 seconds, split into “0-2s, 2-4s, 4-6s”, with the gag landing around the second second.
- The camera. Still for the first two seconds, a fast snap zoom-in on the gag, then a sharp, steady hold. No whip pans, no shake, no motion blur; a few shots use one slow push-in instead.
- Sound. Ambience and effects only. Every prompt ends with “No music. No dialogue: nobody speaks”; we added the music in the edit.
- The look. “Real documentary footage”, natural colours, no logos and no readable text.
The face rule showed up clearly in this video: runs with Mira's face images passed 10 of 24 times, and references without a face passed 6 of 6. The contactless-payment shot got no face image at all, only outfit crops that don't show the face. The stepping-out and selfie shots had been rejected with sets of two and four face images; with the face given only through the outfit close-up, both passed on the first try.
The prompts below are exactly what we sent to the model through the API. The Seedance 2.5 Reference box on Klakix takes up to 2,000 characters, and two of the ten prompts fit as they are. For the others, cut the outfit and car descriptions to one sentence (“@Image2 and @Image3 show her 14th-century outfit; keep it exactly.”): the images already carry the detail, and that saves 250 to 600 characters. You can also trim the brackets in the “Everyone wears…” line and the look line. Turn off “Enhance prompt” in the studio: when it's on, a long prompt is rewritten into a short description and details can drop. Generating all ten shots once at 480p (56 seconds in total) costs 576 credits ($20.92–$38.40), and upscaling the 34-second edit to 1080p costs 43 more. Rejected runs are refunded.
1. The drawbridge: a supercar rolls into the castle · 0.0–4.0 s
Camera: Low hero angle from the bridge deck, 24mm; the camera is locked off and only the car moves.
References: @Image1 car · @Image2 town
Seedance 2.5 Reference · 4 s · 480p · 4:3
@Image1 shows the car: keep it exactly: a bright yellow Italian-style supercar with sharp angular lines, scissor doors, dark tinted windows and black wheels with plain matte-black centre caps; plain bodywork with no logos, no badges, no emblems anywhere (not on the wheels, not on the nose), no text, no licence plate. @Image2 shows the town: match its architecture, pale limestone, red-tiled timber-framed houses, cobbles and warm light. ONE SHOT, 4 seconds, no cuts. Late morning, warm sunlight. The outer end of a lowered wooden drawbridge across a green moat, leading to the town gatehouse: a massive pale limestone gate tower with a raised iron portcullis and heavy iron chains running up into the tower. Two guards in mail shirts and iron kettle helmets, holding halberds, stand on either side of the open gate facing the bridge; a peasant with a handcart of hay has stepped aside at the edge of the bridge. Everyone wears accurate 14th-century clothing (wool tunics, linen coifs, hooded capes; guards in mail shirts and iron kettle helmets). The only modern thing in the whole frame is the yellow supercar, plain and unbranded: no logos, no readable text anywhere. 0-1s: the yellow supercar is already on the drawbridge just in front of the camera, seen from behind at a three-quarter angle, rolling slowly toward the open gate. 1-3s: it rolls on slowly and steadily over the thick planks toward the gate; the chains sway slightly; the two guards do not move, they only follow it with their eyes. 3-4s: the car reaches the shadow of the gate arch; the peasant with the handcart watches it with a blank face. Camera: Heroic low-angle shot from the deck of the drawbridge, the lens just above the planks, 24mm, looking up at the car and the towering gatehouse against the blue sky; the camera is locked off and does not move; only the car moves. The driver is hidden behind the dark tinted rear window. Real documentary footage shot on location, natural true-to-life colours, real skin texture, no CGI look, no colour grading, no film grain, no text on screen. Sound: natural ambience and real sound effects only: a deep supercar engine burble echoing off the stone gate, thick wooden planks creaking under the tyres, iron chains clinking, water birds. No music. No dialogue: nobody speaks; any crowd sound is a wordless murmur or cheer.
2. Stepping out; a stable boy offers hay to the air intake · 4.0–8.5 s
Camera: Wide shot at wheel height, 35mm, apples soft in the foreground; a snap zoom to the groom at 5.5 s, then a sharp hold.
References: @Image1 outfit close-up (face visible) · @Image2 outfit body (faceless crop) · @Image3 outfit detail (faceless crop) · @Image4 car · @Image5 town
Seedance 2.5 Reference · 8 s · 480p · 4:3
@Image1 shows the young woman, Mira: keep her face, freckles, the single small beauty mark on her left cheekbone, hazel-green eyes and her wavy dark-brown hair with curtain bangs exactly. @Image2 and @Image3 show her 14th-century noble outfit; keep it exactly and unchanged: a deep sapphire-blue velvet gown with small gold buttons and gold-embroidered bands, a sapphire velvet mantle with a wide white ermine collar held by a gold chain between two round brooches, a low girdle of gold plaques, a slim gold circlet with a small sapphire, small gold hoop earrings, and flat soft pointed medieval leather shoes with no heels. @Image4 shows the car: keep it exactly: a bright yellow Italian-style supercar with sharp angular lines, scissor doors, dark tinted windows and black wheels with plain matte-black centre caps; plain bodywork with no logos, no badges, no emblems anywhere (not on the wheels, not on the nose), no text, no licence plate. @Image5 shows the town: match its architecture, pale limestone, red-tiled timber-framed houses, cobbles and warm light. ONE SHOT, 8 seconds, no cuts. Late morning, warm sunlight, the cobblestone market square: timber-framed houses, market stalls with baskets of apples and bread, the round stone fountain behind; townsfolk stand at a polite distance. Everyone wears accurate 14th-century clothing (wool tunics, linen coifs, hooded capes; guards in mail shirts and open-faced iron kettle helmets, faces clearly visible, no visors, no face masks). The only modern thing in the whole frame is the yellow supercar, plain and unbranded: no logos, no readable text anywhere. 0-3s: the yellow supercar rolls slowly into the square from the left and comes to a smooth stop on the cobbles, its left side to the camera; the townsfolk step back and watch. 3-5s: the driver's scissor door swings slowly upward and Mira steps out gracefully, lifting the hem of her gown; her flat pointed medieval leather shoe (no heel) touches the cobbles; she stands tall beside the car. 5-8s: a young stable groom in a brown wool tunic hurries in and holds a bundle of hay up to the car's big side air intake, as if feeding a horse; Mira gives him a small deadpan glance. Camera: Static wide shot at wheel height on a 35mm lens, a basket of apples on a market stall softly out of focus in the foreground. 0-5s: the camera is still. At 5.5s a fast snap zoom-in to the groom holding the hay up to the air intake, then hold steady and sharp until the end. No whip pans, no shake. Real documentary footage shot on location, natural true-to-life colours, real skin texture, no CGI look, no colour grading, no film grain, no text on screen. Sound: natural ambience and real sound effects only: the soft hum of the door mechanism, footsteps on cobbles, chickens, the wordless murmur of the market. No music. No dialogue: nobody speaks; any crowd sound is a wordless murmur or cheer.
3. Gold from a cash machine in the castle wall · 8.5–11.7 s
Camera: Over Mira's shoulder, 50mm, her ermine collar soft in front; a snap zoom to the slot at 2 s, then a sharp hold.
References: @Image1 Mira, front portrait · @Image2 Mira, 3/4 close-up · @Image3 outfit body (faceless crop) · @Image4 outfit detail (faceless crop) · @Image5 town
Seedance 2.5 Reference · 6 s · 480p · 4:3
@Image1 and @Image2 show the same young woman, Mira: keep her face, freckles, the single small beauty mark on her left cheekbone, hazel-green eyes and her wavy dark-brown hair with curtain bangs exactly. @Image3 and @Image4 show her 14th-century noble outfit; keep it exactly and unchanged: a deep sapphire-blue velvet gown with small gold buttons and gold-embroidered bands, a sapphire velvet mantle with a wide white ermine collar held by a gold chain between two round brooches, a low girdle of gold plaques, a slim gold circlet with a small sapphire, small gold hoop earrings, and flat soft pointed medieval leather shoes with no heels. @Image5 shows the town: match its architecture, pale limestone, red-tiled timber-framed houses, cobbles and warm light. ONE SHOT, 6 seconds, no cuts. Midday, warm sunlight. Beside the town gate, a modern cash machine is set into a carved pointed-arch stone niche in the pale limestone wall: a brushed-steel front, a softly glowing plain blue screen, a plain keypad and a wide slot. Right beside the niche, a knight in full steel plate armour stands with both gauntlets resting on the pommel of his sword, like a bank security guard. Everyone wears accurate 14th-century clothing (wool tunics, linen coifs, hooded capes; guards in mail shirts and iron kettle helmets). The only modern thing in the whole frame is the cash machine (its screen shows only a plain glow: no text, no numbers, no icons), plain and unbranded: no logos, no readable text anywhere. 0-2s: Mira holds her embroidered leather purse open under the slot; the machine whirrs; the knight stands guard beside it. 2-4s: a thick, glittering cascade of gold coins pours from the slot into the purse, coins tumbling and flashing in the sun. 4-6s: the purse is full and heavy; she pulls the drawstring tight while the coins keep glinting. Camera: Over-the-shoulder medium shot from just behind Mira's left shoulder on a 50mm lens: her ermine collar and hair softly out of focus in the foreground, the cash machine sharp in its stone niche, the armoured knight beside it. 0-2s: the camera is still. At 2s a fast snap zoom-in to the coin slot and the purse, then hold steady and sharp on the pouring gold coins until the end. No whip pans, no shake, no motion blur on the coins. Real documentary footage shot on location, natural true-to-life colours, real skin texture, no CGI look, no colour grading, no film grain, no text on screen. Sound: natural ambience and real sound effects only: the soft whirr of the machine, a long cascade of clinking gold coins, birds, a distant wordless market murmur. No music. No dialogue: nobody speaks; any crowd sound is a wordless murmur or cheer.
4. Contactless payment at the blacksmith's · 11.7–14.3 s
Camera: Over the blacksmith's shoulder, 50mm; a snap zoom to the card at 2 s, then a hold.
References: @Image1 outfit body (faceless crop) · @Image2 outfit detail (faceless crop)
Seedance 2.5 Reference · 6 s · 480p · 4:3
@Image1 and @Image2 show a noblewoman's 14th-century outfit; keep its sleeve exactly: deep sapphire-blue velvet with small gold buttons from elbow to wrist and a gold-embroidered cuff, gold rings on her fingers. ONE SHOT, 6 seconds, no cuts. Midday. A blacksmith's open-fronted forge on the market square, under a timber roof: bright daylight from the street plus the orange glow of the hearth; a glowing horseshoe on the anvil; sparks drift in the soft background. Everyone wears accurate 14th-century clothing (wool tunics, linen coifs, hooded capes; guards in mail shirts and iron kettle helmets). The only modern thing in the whole frame is a small black handheld card payment terminal (its little screen shows only a plain glow: no text, no numbers), plain and unbranded: no logos, no readable text anywhere. 0-2s: the blacksmith holds the card terminal out over the anvil; a slender woman's hand with gold rings, in the sapphire velvet sleeve with gold buttons, brings a completely plain matte-black card, blank on both sides, toward it. 2-4s: the card touches the terminal and a small green light blinks with a soft beep. 4-6s: she withdraws her hand and the blacksmith picks up his hammer again. Camera: Over-the-shoulder medium shot from behind the blacksmith's right shoulder on a 50mm lens: his scorched leather apron strap and shoulder softly out of focus in the foreground, the terminal and her hand sharp. 0-2s: the camera is still. At 2s a fast snap zoom-in to the card touching the terminal, then hold steady and sharp until the end. No whip pans, no shake, no motion blur. Real documentary footage shot on location, natural true-to-life colours, real skin texture, no CGI look, no colour grading, no film grain, no text on screen. Sound: natural ambience and real sound effects only: the crackle of the forge fire, a soft electronic beep, a hammer ringing on an anvil. No music. No dialogue: nobody speaks; any crowd sound is a wordless murmur or cheer.
5. The drift: spinning tyre and smoke · 14.3–16.8 s
Camera: Low medium shot at cobblestone level, 50mm; a snap zoom to the tyre at 1.5 s, one slow pull-back at 4 s.
References: @Image1 car · @Image2 town
Seedance 2.5 Reference · 6 s · 480p · 4:3
@Image1 shows the car: keep it exactly: a bright yellow Italian-style supercar with sharp angular lines, scissor doors, dark tinted windows and black wheels with plain matte-black centre caps; plain bodywork with no logos, no badges, no emblems anywhere (not on the wheels, not on the nose), no text, no licence plate. @Image2 shows the town: match its architecture, pale limestone, red-tiled timber-framed houses, cobbles and warm light. ONE SHOT, 6 seconds, no cuts. Afternoon, bright warm sunlight, the cobblestone market square: the round stone fountain, timber-framed houses with red-tiled roofs and market stalls; townsfolk stand safely back along the house fronts and watch. Everyone wears accurate 14th-century clothing (wool tunics, linen coifs, hooded capes; guards in mail shirts and iron kettle helmets). The only modern thing in the whole frame is the yellow supercar, seen only from its left side and rear; its nose is never visible, plain and unbranded: no logos, no readable text anywhere. 0-1.5s: the yellow supercar slides sideways past the fountain in a smooth, controlled drift, seen from its left side and rear; its rear tyre spins on the cobbles. 1.5-4s: thick white smoke pours from the spinning rear tyre and billows up, glowing in the sunlight. 4-6s: the car keeps circling the fountain inside its own cloud of smoke; the townsfolk watch with blank faces. Camera: Low-angle medium shot at cobblestone level on a 50mm lens, the corner of a market stall's canvas awning softly out of focus in the foreground. 0-1.5s: the camera is still. At 1.5s a fast snap zoom-in to the spinning rear tyre and the smoke pouring from it, then hold steady and sharp on the tyre and the smoke. At 4s one slow pull-back reveals the fountain and the watching crowd. No whip pans, no shake, no motion blur on the car body. Real documentary footage shot on location, natural true-to-life colours, real skin texture, no CGI look, no colour grading, no film grain, no text on screen. Sound: natural ambience and real sound effects only: a roaring supercar engine, screeching tyres, a wordless gasp from the crowd. No music. No dialogue: nobody speaks; any crowd sound is a wordless murmur or cheer.
6. Top-down drift: a smoke ring around the fountain · 16.8–19.4 s
Camera: Straight-down drone shot; the camera never moves.
References: @Image1 car · @Image2 town
Seedance 2.5 Reference · 4 s · 480p · 4:3
@Image1 shows the car: keep it exactly: a bright yellow Italian-style supercar with sharp angular lines, scissor doors, dark tinted windows and black wheels with plain matte-black centre caps; plain bodywork with no logos, no badges, no emblems anywhere (not on the wheels, not on the nose), no text, no licence plate. @Image2 shows the town: match its architecture, pale limestone, red-tiled timber-framed houses, cobbles and warm light. ONE SHOT, 4 seconds, no cuts. Afternoon, bright sunlight. A straight-down overhead drone view of the cobblestone market square: the round stone fountain in the centre, market stalls with canvas awnings and red-tiled roofs at the edges of the frame, townsfolk standing safely back along the house fronts, a few chickens. Everyone wears accurate 14th-century clothing (wool tunics, linen coifs, hooded capes; guards in mail shirts and iron kettle helmets). The only modern thing in the whole frame is the yellow supercar, plain and unbranded: no logos, no readable text anywhere. 0-1s: the yellow supercar is already sliding sideways around the fountain in a smooth, controlled drift. 1-3s: it keeps circling the fountain sideways, drawing a ring of white tyre smoke on the cobbles; chickens flutter out of the way; the crowd stays back at the edges. 3-4s: a perfect smoke ring hangs around the fountain as the car completes the circle. Camera: Static top-down overhead shot from a high drone, looking straight down; the camera does not move at all; only the car and the smoke move. Real documentary footage shot on location, natural true-to-life colours, real skin texture, no CGI look, no colour grading, no film grain, no text on screen. Sound: natural ambience and real sound effects only: a revving supercar engine, screeching tyres, squawking chickens, a wordless gasp from the crowd. No music. No dialogue: nobody speaks; any crowd sound is a wordless murmur or cheer.
7. A money gun over the crowd from the castle stairs · 19.4–23.3 s
Camera: Slow motion; low angle from inside the crowd, 35mm, one slow push-in.
References: @Image1 Mira, front portrait · @Image2 Mira, 3/4 close-up · @Image3 outfit, full length · @Image4 outfit close-up (face visible) · @Image5 town
Seedance 2.5 Reference · 4 s · 480p · 4:3
@Image1 and @Image2 show the same young woman, Mira: keep her face, freckles, the single small beauty mark on her left cheekbone, hazel-green eyes and her wavy dark-brown hair with curtain bangs exactly. @Image3 and @Image4 show her 14th-century noble outfit; keep it exactly and unchanged: a deep sapphire-blue velvet gown with small gold buttons and gold-embroidered bands, a sapphire velvet mantle with a wide white ermine collar held by a gold chain between two round brooches, a low girdle of gold plaques, a slim gold circlet with a small sapphire, small gold hoop earrings, and flat soft pointed medieval leather shoes with no heels. @Image5 shows the town: match its architecture, pale limestone, red-tiled timber-framed houses, cobbles and warm light. ONE SHOT, 4 seconds, no cuts. Late afternoon, the low golden sun behind her. Mira stands at the top of the broad outer stone staircase of the castle keep; a crowd of peasants in wool tunics and linen coifs fills the courtyard at the foot of the stairs. Everyone wears accurate 14th-century clothing (wool tunics, linen coifs, hooded capes; guards in mail shirts and iron kettle helmets). The only modern thing in the whole frame is a plain matte-black handheld money gun (a party cash cannon) in her hand, which sprays plain blank pale-gold paper notes with no print, no text, no numbers and no faces, plain and unbranded: no logos, no readable text anywhere. 0-1s: she holds the money gun up over the crowd, notes already flying out of it. 1-3s: in slow motion the notes stream out and drift down through the golden backlight like falling leaves; the peasants' raised hands reach up in the foreground. 3-4s: she keeps spraying with a calm, satisfied smile while the notes keep floating down. Camera: Filmed in slow motion. Low angle from inside the crowd at the foot of the stairs, 35mm, looking up at Mira against the bright sky; one slow, steady push-in; the sun rims her hair and the ermine collar, and light bouncing off the pale stone steps lights her face clearly. Real documentary footage shot on location, natural true-to-life colours, real skin texture, no CGI look, no colour grading, no film grain, no text on screen. Sound: natural ambience and real sound effects only: the rhythmic whirr of the money gun, fluttering paper, a joyful wordless cheer from the crowd. No music. No dialogue: nobody speaks; any crowd sound is a wordless murmur or cheer.
8. Champagne with a sparkler at the feast · 23.3–25.8 s
Camera: Symmetrical one-point shot down the table, 28mm; one slow dolly in, no zoom.
References: @Image1 Mira, front portrait · @Image2 Mira, 3/4 close-up · @Image3 outfit, full length · @Image4 outfit close-up (face visible)
Seedance 2.5 Reference · 4 s · 480p · 4:3
@Image1 and @Image2 show the same young woman, Mira: keep her face, freckles, the single small beauty mark on her left cheekbone, hazel-green eyes and her wavy dark-brown hair with curtain bangs exactly. @Image3 and @Image4 show her 14th-century noble outfit; keep it exactly and unchanged: a deep sapphire-blue velvet gown with small gold buttons and gold-embroidered bands, a sapphire velvet mantle with a wide white ermine collar held by a gold chain between two round brooches, a low girdle of gold plaques, a slim gold circlet with a small sapphire, small gold hoop earrings, and flat soft pointed medieval leather shoes with no heels. ONE SHOT, 4 seconds, no cuts. Late afternoon, the great hall of the castle: one long oak banquet table runs straight down the middle of the hall from the camera to the far wall, covered with white linen, silver platters of roast meat, bread, grapes and pewter goblets; tall arched windows on both sides flood the hall with bright daylight and dozens of candles burn on iron wheel chandeliers; nobles in rich 14th-century clothes sit along both sides. At the far end, in the centre of the frame, Mira sits on a high-backed carved oak chair. Everyone wears accurate 14th-century clothing (wool tunics, linen coifs, hooded capes; guards in mail shirts and iron kettle helmets). The only modern thing in the whole frame is a champagne bottle with a fizzing sparkler on its neck (plain gold foil, no label text), plain and unbranded: no logos, no readable text anywhere. 0-1s: the lively feast; Mira sits calmly at the head of the table. 1-3s: a servant in a livery tunic walks up beside the table toward her, holding the champagne bottle high, its sparkler spraying bright sparks; the nobles turn their heads to watch it pass. 3-4s: he sets the bottle down in front of Mira; she gives a tiny satisfied nod. Camera: Perfectly symmetrical one-point-perspective shot down the length of the table from its near end, at the eye level of the seated guests, 28mm; one slow, steady dolly push-in along the table toward Mira; no zoom, nothing fast. The hall is bright and evenly lit; every face is clearly visible. Real documentary footage shot on location, natural true-to-life colours, real skin texture, no CGI look, no colour grading, no film grain, no text on screen. Sound: natural ambience and real sound effects only: the fizz and crackle of the sparkler, clinking goblets, a lively wordless murmur of the guests. No music. No dialogue: nobody speaks; any crowd sound is a wordless murmur or cheer.
9. Back from shopping on a dragon · 25.8–29.7 s
Camera: Low wide shot from the courtyard floor, 35mm, the dragon's tail soft in front; a snap zoom at 3.5 s, then a hold.
References: @Image1 Mira, front portrait · @Image2 Mira, 3/4 close-up · @Image3 outfit body (faceless crop) · @Image4 outfit detail (faceless crop) · @Image5 dragon · @Image6 town
Seedance 2.5 Reference · 8 s · 480p · 4:3
@Image1 and @Image2 show the same young woman, Mira: keep her face, freckles, the single small beauty mark on her left cheekbone, hazel-green eyes and her wavy dark-brown hair with curtain bangs exactly. @Image3 and @Image4 show her 14th-century noble outfit; keep it exactly and unchanged: a deep sapphire-blue velvet gown with small gold buttons and gold-embroidered bands, a sapphire velvet mantle with a wide white ermine collar held by a gold chain between two round brooches, a low girdle of gold plaques, a slim gold circlet with a small sapphire, small gold hoop earrings, and flat soft pointed medieval leather shoes with no heels. @Image5 shows Mira's dragon: keep it exactly: a huge, friendly bronze-gold dragon with cream belly plates, folded leathery wings, two curved horns, calm amber eyes and a gentle, dog-like expression, a brown leather saddle with gold trim on its back. @Image6 shows the town: match its architecture, pale limestone, red-tiled timber-framed houses, cobbles and warm light. ONE SHOT, 8 seconds, no cuts. Late afternoon, warm golden sunlight. The castle's inner courtyard: pale limestone walls and towers, a wooden gallery, a stone well; knights in plate armour with open-faced helmets and a few townsfolk stand back along the walls. Mira's dragon has just landed on the cobbles, wings half folded. Everyone wears accurate 14th-century clothing (wool tunics, linen coifs, hooded capes; guards in mail shirts and open-faced iron kettle helmets, faces clearly visible, no visors, no face masks). The only modern thing in the whole frame is a bunch of six glossy paper shopping bags in cream and black with rope handles that Mira carries, plain and unbranded, plain and unbranded: no logos, no readable text anywhere. 0-3s: the dragon lowers its shoulder to the cobbles like a kneeling horse; Mira sits in the saddle holding the shopping bags. 3-5s: she slides gracefully down the dragon's shoulder onto the cobbles with all the bags in her hands. 5-8s: she walks toward the camera with a calm, satisfied smile while the dragon behind her lowers its head and wags its tail like a well-behaved dog; a knight bows. Camera: Static wide low-angle shot from the courtyard floor on a 35mm lens, the tip of the dragon's tail softly out of focus in the foreground. 0-3.5s: the camera is still. At 3.5s a fast snap zoom-in to Mira and the shopping bags, then hold steady and sharp until the end. No whip pans, no shake, no motion blur on her face. Real documentary footage shot on location, natural true-to-life colours, real skin texture, no CGI look, no colour grading, no film grain, no text on screen. Sound: natural ambience and real sound effects only: the dragon's deep, calm breathing, leathery wings settling, rustling paper bags, footsteps on cobbles, a wordless murmur. No music. No dialogue: nobody speaks; any crowd sound is a wordless murmur or cheer.
10. A rampart selfie with the dragon · 29.7–33.9 s
Camera: Selfie POV on a phone's 24mm at arm's length; a snap zoom at 2.5 s, then a sharp hold.
References: @Image1 outfit close-up (face visible) · @Image2 outfit body (faceless crop) · @Image3 outfit detail (faceless crop) · @Image4 dragon · @Image5 car · @Image6 town
Seedance 2.5 Reference · 6 s · 480p · 4:3
@Image1 shows the young woman, Mira: keep her face, freckles, the single small beauty mark on her left cheekbone, hazel-green eyes and her wavy dark-brown hair with curtain bangs exactly. @Image2 and @Image3 show her 14th-century noble outfit; keep it exactly and unchanged: a deep sapphire-blue velvet gown with small gold buttons and gold-embroidered bands, a sapphire velvet mantle with a wide white ermine collar held by a gold chain between two round brooches, a low girdle of gold plaques, a slim gold circlet with a small sapphire, small gold hoop earrings, and flat soft pointed medieval leather shoes with no heels. @Image4 shows Mira's dragon: keep it exactly: a huge, friendly bronze-gold dragon with cream belly plates, folded leathery wings, two curved horns, calm amber eyes and a gentle, dog-like expression, a brown leather saddle with gold trim on its back. @Image5 shows the car: keep it exactly: a bright yellow Italian-style supercar with sharp angular lines, scissor doors, dark tinted windows and black wheels with plain matte-black centre caps; plain bodywork with no logos, no badges, no emblems anywhere (not on the wheels, not on the nose), no text, no licence plate. @Image6 shows the town: match its architecture, pale limestone, red-tiled timber-framed houses, cobbles and warm light. ONE SHOT, 6 seconds, no cuts. Golden hour at sunset. Mira stands on the walkway of the castle rampart beside a stone crenellation; far below her lie the town's red roofs, the gatehouse and the lowered drawbridge, and at the foot of the drawbridge the parked yellow supercar. Everyone wears accurate 14th-century clothing (wool tunics, linen coifs, hooded capes; guards in mail shirts and open-faced iron kettle helmets, faces clearly visible, no visors, no face masks). The only modern thing in the whole frame is the yellow car far below, plain and unbranded: no logos, no readable text anywhere. 0-2s: Mira smiles into the lens and gives one slow, confident wink. 2-4s: behind her, the dragon's huge head rises slowly over the crenellations into the frame and leans in next to her face, as if posing for the selfie. 4-6s: Mira laughs softly, rests her cheek against the dragon's jaw and blows a small kiss to the lens; the dragon blinks calmly. Camera: Selfie POV on a phone's 24mm lens at arm's length, eye level, no phone visible; steady, tiny natural hand movement. 0-2.5s: still. At 2.5s a quick snap zoom-in framing Mira and the dragon's head together, then hold steady and sharp. No whip pans, no shake. Low sun to her side: warm key light, subtle natural lens flare. Real documentary footage shot on location, natural true-to-life colours, real skin texture, no CGI look, no colour grading, no film grain, no text on screen. Sound: natural ambience and real sound effects only: evening wind, the dragon's deep calm breath, swallows, a soft wordless laugh. No music. No dialogue: nobody speaks; any crowd sound is a wordless murmur or cheer.
What it costs on Klakix
Credit prices come from our catalog; the dollar range depends on the credit pack you buy.
| Step | What you get | Credits |
|---|---|---|
| Character | 4 candidate portraits with GPT Image 2.5, plus the 3/4 and full-body angles of the one you save | 35 ($1.27–$2.33) |
| New scene | The saved character drawn in a new setting | 8 ($0.29–$0.53) |
| Test clip | Seedance 2.5 Reference, 5 seconds, 480p | 52 ($1.89–$3.47) |
| Long take | Seedance 2.5 Reference, 30 seconds, 480p, upscaled to 1080p | 346 ($12.57–$23.07) |
| Same take at 720p | Seedance 2.5 Reference, 30 seconds, 720p, for comparison | 692 ($25.13–$46.13) |
| Voice | 1,000 characters with ElevenLabs v4 | 10 ($0.36–$0.67) |
Signing up is free. Image and video tools open after your first credit pack, which starts at $5 for 75 credits, and failed generations are refunded automatically. All packs are on the pricing page.
Common questions
How do I make an AI influencer?
Write a fixed description of a fictional adult, generate a few portraits and keep one as the anchor, then make every other image and video with that anchor as the reference. Give the character one voice, post on a schedule, and turn on the platform's AI labels. The face stays consistent because everything is built from the same anchor.
How much does it cost to create an AI influencer?
On Klakix, the character itself (4 candidate portraits plus the 3/4 and full-body angles of the one you keep) is 35 credits with GPT Image 2.5 or 48 with Nano Banana 2. A 5-second Seedance 2.5 Reference test clip at 480p is 52 credits, and a 30-second take at 480p with a 1080p upscale is 346 credits. Credit packs start at $5.
Can I make an AI influencer for free?
Not on Klakix: every image and video is made by a paid AI model, so they use credits. Signing up is free, and the 20 credits you get after verifying your email work for voiceovers; image and video tools open after your first credit pack, which starts at $5 for 75 credits.
How do I keep my AI influencer's face consistent?
Use one anchor portrait as the reference for everything, generate each angle as its own high-resolution image instead of one multi-panel sheet, and give a video model the single angle images. Pick one identity mark, like a beauty mark on one cheek, and check it on every image: if it moved, the image was mirrored.
Is it legal to create an AI influencer?
Creating and posting a fictional AI person is generally legal, but the rules around it still apply: label the account and realistic posts as AI, disclose paid posts, and never use a real person's likeness without permission. In the US, the FTC's rule on fake reviews, in force since October 21, 2024, bans testimonials that misrepresent that the reviewer exists, AI-generated ones included. This isn't legal advice.
Do AI influencers make money?
Some do, through the same routes as human creators: brand deals, affiliate links and platform programs. We haven't found earnings figures we could verify, so we don't quote any, and no tool can promise followers or income.
How do I make an AI influencer on Instagram?
Create a separate account for the character, say in the bio that the person is AI-generated, and turn on the AI-generated profile label from Edit profile. Instagram says labeled accounts keep their reach, while accounts that feature an AI person without the label can lose recommendations. Label realistic AI posts too.
Why does Seedance reject my AI character?
Seedance rejects photos of real people. Realistic AI-generated characters given as single images often pass, but not every time: in our November 2–3, 2026 tests, runs with face images passed 18 of 32 times and references without a face passed 6 of 6. It also rejected a multi-panel character sheet once and a product photo with a real model in it. Give it the single angle images; if a run is rejected, cut the face images down to one, give the outfit as crops that don't show the face, or send it again. If the filter blocks a generation here, the credits come back automatically.
Which AI is best for making an AI influencer?
For Mira's images we used GPT Image 2.5, which ranked first for both text to image and image editing on the Artificial Analysis leaderboards in November 2026. For video with sound we used Seedance 2.5, ranked first for image to video with audio on the same leaderboards that month. Rankings change, so check them before a big project.
Create your AI character
Pick the look from menus, keep one of 4 portraits, and reuse the same face in your videos. Credit packs start at $5.
Create a character