If you make comics, webtoons or short drama series with AI, the usual problem is a character who won't stay the same. One face in the first panel, a slightly different one in the next, and the jacket changes colour along the way.
The fix is a character reference sheet. Put the front, side and back of the character on one image, then use that image as a reference for every panel after it. Now the model has something to copy.
This guide starts with the simplest case and adds one thing at a time: a character sheet from one photo, a four-view character turnaround from text only, one character built from three images, then outfits and props as separate images, and last a pose image for the final scene. Every step comes with its prompt.
What you need
- Open the CVY.AI image generator. Use Text to Image when you only have words, and Image to Image when you upload images.
- You can add up to 10 reference images. "Image 1, image 2" in a prompt means the reference images from left to right. Drag them to change the order.
- Model: the first three steps used GPT Image 2 at 2K. The two biker steps used GPT Image 2.6 Sunburst.
- Set the Aspect Ratio named in each step.
One note on the prompts: we wrote the prompt in step 3 in English and show it as it was. The others were written in Chinese, and what you see here is a translation.
Step 1: A character reference sheet from one photo
You need one settled picture of the character first. It can be a character you drew, or one you generate from text. Ours is a half-body portrait generated from text.
Choose Image to Image, add the picture as a reference image, set the aspect ratio to 16:9, and use this prompt:
Use the uploaded photo as the only reference for the person and create a character reference sheet of the same person. On the left, three full-body views in this order: front, side, back, all in the same upright standing pose with arms relaxed at the sides. On the right, face close-ups: one straight on and one three-quarter view. Keep the face shape, features, hairstyle, hair colour and clothing the same as in the reference photo. Pure white background, soft even studio light, realistic photo look, high detail. Only this one person in the image. No text, labels, arrows, props or scenery.
The face, hair and sweater match the reference photo. At 16:9, one row has room for three full-body views and two close-ups.
One thing to know: our reference photo only shows the upper body, so the model chose the jeans and shoes itself. If the lower half matters, use a full-body reference or describe it in the prompt.
The same prompt, word for word, also works with a photo of a man.
Step 2: No photo? A four-view character turnaround from text
If the character only exists in your head, you can skip the reference image and describe them. For vertical formats, 9:16 works well: two rows on one image, with front, left side, right side and back.
Choose Text to Image, set the aspect ratio to 9:16, and use this prompt:
Four-view character turnaround, vertical layout, two rows with two views each: front, left side, right side, back. The same character in the same upright standing pose, the same height and proportions, with clothing and hairstyle exactly the same in all four views. Character: [character description]. Pure white background, soft even studio light, [art style], high detail. No text, labels, arrows or scenery.
To change the character, rewrite only the sentence after "Character:". Change the art style to suit the story. The two sheets below used:
- A 24-year-old swordswoman in a Chinese period drama, high ponytail, pale blue and white fitted robes, a sword at her waist. Art style: realistic costume-test photo for a TV drama.
- A 26-year-old office worker, low ponytail, beige blazer with a white shirt and matching trousers. Art style: realistic photo.
Step 3: Set the face, body and art style separately
So far, everything about the character sits in one photo or one sentence. If you want a different build, or a different art style, you start over.
So split them up: one image for the face, one for the body, and one that only lends its art style. Upload all three and tell the model what to take from each image, and what to leave.
Choose Image to Image, add the face, body and style images in that order, set the aspect ratio to 16:9, and use this prompt:
Create a character turnaround sheet for one single character, combining the reference images: the face, hair, eyes, stubble and earrings from image 1; the body build, muscles, skin tone and outfit from image 2 (shirtless, loose black jogger sweatpants with a white drawstring, white sneakers); the art style, line art, shading and lighting from image 3. Image 3 is only an art style reference: ignore its face, hair, clothing and pose, and copy only its line art, shading and lighting.
Layout: exactly four full-body views of the same man in one row, evenly spaced, all the same height and scale, feet on the same ground line, on a plain light grey background. Nothing else in the image: no face close-up, no inset, no extra figures.
View 1, FRONT: body and face facing the viewer straight on, both shoulders equally wide, both eyes and both ears visible.
View 2, 3/4: the whole body rotated 45 degrees to the viewer's left, the far shoulder clearly narrower than the near one, the far side of the chest partly hidden, face turned the same way with the far eye partly hidden by the nose bridge.
View 3, SIDE: a true 90-degree left profile of the whole body, facing the left edge of the image, only one arm, one eye and one ear visible, the chest and abs seen only as a side silhouette.
View 4, BACK: body and head facing directly away from the viewer, back muscles, shoulder blades and back of the head visible, no face visible.
Neutral standing pose in every view, arms relaxed at the sides. The face, hairstyle, body proportions and outfit must stay exactly the same in all four views. Below each view, a small clean label in a simple sans-serif font: FRONT, 3/4, SIDE, BACK. No other text, no watermark. Hands with five fingers and correct anatomy. Art style: modern Korean webtoon (manhwa) illustration, clean thin line art, soft cel shading, even soft lighting, high detail.
Two habits do most of the work here:
- For every image, say what to take and what to ignore. The style image has a person in it too. Without "ignore its face, hair, clothing and pose", his looks can leak into your character.
- Give each view its own sentence, and say what can be seen from that angle. For the side view, "only one arm, one eye and one ear visible". For the back view, "no face visible".
Step 4: Split two more things, the outfit and the prop
The same idea goes further. We made a biker character from four kinds of images: face, body, riding suit and motorcycle. Make a few of each, then pick one of each when you need a character.
All of them come from text. Each kind has its own template, and they share one ending. The art style lives in that ending, so the pieces still look like one set when you combine them:
Plain light grey background, no text and no watermark, no brand logos and no licence plates anywhere. Art style: Japanese anime, clean line art, cel shading, even soft lighting, high detail.
One example of each kind:
- Face, 1:1: Head-and-shoulders portrait, facing the camera, a 25-year-old East Asian man with short, slightly messy black hair, sharp eyes and brows, a cool expression, looking at the camera.
- Body, 3:4: Full-body front view, standing with arms relaxed at the sides, an adult man 185 cm tall, slim and lean with broad shoulders and long legs, wearing a plain black fitted long-sleeve top and trousers and black trainers, and a plain black full-face helmet with a dark visor, face not visible.
- Riding suit, 3:4: A one-piece leather track suit, mostly matte black, with orange stripes on the shoulders, the outside of the arms and the lower legs, hard armour on the shoulders and elbows, knee sliders on the outside of the knees, with black gloves and black racing boots, worn on a headless grey mannequin, full front view from shoulders to feet.
- Motorcycle, 4:3: A fully faired sports bike in matte black with thin orange lines, full side view, front of the bike pointing to the left of the image, parked on the ground, no people in the image.
The body wears a helmet on purpose, and the suit sits on a mannequin with no head. That way there is only one face among the references, and no other face can slip into the result.
Then add four images in this order: face, body, riding suit, motorcycle. Set the aspect ratio to 4:3 and use this prompt:
Use the 4 uploaded reference images to create a character card for one character. Take the face, hairstyle, eyes and expression from image 1. From image 2 take only the height, build and body proportions; ignore the helmet and the clothes. From image 3 take the cut, colours, stripes and armour of the riding suit; ignore the mannequin. From image 4 take the model, colours and details of the motorcycle. Layout: on the left, his full-body front view and full-body back view, standing; top centre, a head-and-shoulders close-up; on the right, a full side view of his motorcycle with a matte black full-face helmet on the ground next to it. Only this one character and this one motorcycle in the image. Hands with five fingers.
Add the shared ending after it.
Swap in another face, another outfit and another bike, leave the prompt alone, and you have a different character:
The small features on each face came through: a pointed canine tooth, short stubble, a scar on the brow. The stripes on the suits and the colours of the bikes match as well.
Step 5: Add a pose image for the final scene
Everyone on a character card is standing still. In an actual panel he leans into a corner or crouches down to check the bike, and a pose is hard to pin down in words.
So we add one more image: a grey figure in the pose we want. It also comes from text, at 3:4:
Pose reference: a grey 3D figure with no facial features and no clothing details, crouching beside the rear wheel of a plain grey motorcycle, one hand on the tyre, looking at the chain. Plain light grey background, simple 3D render, no text.
Then add five images in this order: face, body, riding suit, motorcycle, pose. Set the aspect ratio to 3:4 and use this prompt:
Use the 5 uploaded reference images to create one scene. Take the face, hairstyle and expression from image 1. From image 2 take only the height, build and body proportions; ignore the helmet and the clothes. From image 3 take the cut, colours, stripes and armour of the riding suit; ignore the mannequin. From image 4 take the model, colours and details of the motorcycle. From image 5 take only the pose and action, and where the person is in relation to the motorcycle; ignore the colour and material of the figure and the grey motorcycle. Scene: a garage, a wall of tools, warm yellow light. Hands with five fingers. No text and no watermark, no brand logos and no licence plates anywhere. Art style: Japanese anime, clean line art, cel shading, even soft lighting, high detail.
The action followed the pose image, and so did the positions of rider and bike. This one also uses a new mix: the brown-haired face with the black and orange suit and the sports bike. The pieces are separate, so you can combine them any way you like.
To change the setting, rewrite only the sentence after "Scene:". To change the action, swap the pose image:
A few lessons
- Faces hold best. Hair, features, and small things like a pointed tooth or a scar carried over every time we tried.
- Outfits and props hold too. Cut, colours and stripes match, as long as each one has its own clean image.
- Side views need one more sentence. With only "90-degree profile", the head often turns most of the way and the far eye still shows. Adding "only one eye visible, the far eye and cheek not visible" helps a lot.
FAQ
What is a character reference sheet?
One image that shows the same character from several angles, usually front, side and back, often with a face close-up. Artists use it to keep a character the same from drawing to drawing. With AI you use it the same way: as the reference image for every new picture of that character.
How do I write a prompt for a character reference sheet?
Cover four things: which views and in what order; the same person in the same pose, height and proportions; a plain background with even light; no text, labels or scenery. If you have a reference photo, add "use the uploaded photo as the only reference for the person".
What is the difference between a character reference sheet and a character turnaround?
A turnaround shows the full body rotating through fixed angles: front, 3/4, side, back. A reference sheet is the wider term and can also hold close-ups, outfits and props. Use four views when the character is not the same on both sides, for example a scar on one cheek or a bag on one shoulder.
Can I use a photo of a real person?
Yes, the method is the same as step 1. Use only your own photo, or one you have permission to use.
Ready to try it? Open the CVY.AI image generator, switch to Image to Image, and add your reference images in order.








