Does the character look great in the first frame, but in the second has a different nose, hair, or jacket? Before you create a series of images or a film, prepare a reference card: several organized views of the same person, a description of features, and a list of elements that must not change. We will show ready prompts, the process of transitioning from card to video, and other cards useful for products, locations, and entire campaigns.
Status as of October 2026 (availability of the mentioned models checked on the service pages on October 23, 2026). Illustrations are original examples. A reference card increases the chance of consistency, but no current generator guarantees the same face in every scene.
What is a character card
It is a working document for subsequent generations. Instead of writing "woman in a jacket" every time, you show the model a specific person: a front and side view of the face, silhouette, clothing, important details, expression, and colors. The card also helps a person notice errors: if a scar suddenly changes the eyebrow, you immediately know what to correct.

First, choose a good source image
The best reference shows the face, hairstyle, and most of the outfit in even lighting. Sharp sunglasses, a hand covering the cheek, or a very dark filter will make it difficult to recreate features. If you only have a portrait, do not expect a faithful depiction of the shoes: the model will have to invent them. Note this uncertainty on the card.
Our character has a medium complexion with a copper tone, a short silver undercut, a small scar crossing the right eyebrow, a mustard field jacket, a navy scarf, and dark pants. These are features that can be visually checked. The term "charismatic" may help with the mood, but does not serve as a visual block.
Break the card into readable parts
| Part | Purpose | What to check |
|---|---|---|
| Profile | One-sentence description and fixed features | Are there any features added that are not visible in the photo? |
| Body rotations | Front, three-quarters, side, back | Height and length of clothing in the same scale. |
| Close-ups of the face | Geometry of the face and hairstyle | Eyebrows, nose line, scar, skin tone. |
| Expression | Reactions without changing identity | The same face also when smiling and surprised. |
| Outfit | Seams, layers, fasteners, shoes | The same clothing parts in all panels. |
| Colors | Palette for later scenes | Samples after manually reading the color from the image. |
A large board with everything at once can be impressive, but small faces and text may be incorrect. In practice, it is better to create separate rotations and expressions, and only then compile the approved images into one document. Do not treat HEX codes generated as text on graphics as measurements; download the color with a color picker in the editor.


How to prepare a card in Picsart
Open Picsart AI Playground. In October 2026, the GPT Image 2 model is on the list of this service; if you have access to it in your plan, select it, go to editing with the image, and add your own reference. The names of modes and limits depend on the current panel. First, generate rotations, then expressions. Choose the best results and keep the source image next to them.
Use the attached image as the sole source of identity for the fictional heroine. Prepare four full views of the same person: from the front, at a three-quarters angle, from the side, and from the back. All views in the same scale, neutral posture, whole shoes and top of the head visible, uniform light background, soft studio light.
Keep the same face structure, skin tone, short silver hairstyle, scar on the right eyebrow, mustard jacket, navy scarf, dark pants, and shoes. Show the logical construction of the outfit from the back and side. No alternative clothing, new accessories, other characters, or text. If a detail is not visible in the reference, fill it in as sparingly as possible.
Based on the attached reference, prepare a grid of four portraits from the shoulders up. This is exactly the same fictional heroine and the same outfit in each panel. Emotions: neutral, mildly amused, worried, determined. Change the muscles of the face, not its structure. The same scar on the right eyebrow, eye color, hairline, skin tone, and lighting. Bright neutral background, no signatures or additional people.
Variant for the series: eight emotions and six poses
Four portraits and several views of the silhouette are enough for the first short shot. If the heroine is to return in many scenes, expand the material only when you know what reactions and movements the script requires. For example: a calm conversation requires glances and smiles, while a discovery scene may need surprise, worry, and determination. Do not generate dozens of poses just to fill the board.
For our heroine, we prepared two separate sheets. The first shows eight changes in expression, the second six body positions. This is still material for selection, not a promise that the video model will replicate every expression or gesture. In a large grid, faces and clothing details are small, so for the proper shot, choose the approved panel and save it as a separate image.

Use the attached approved portrait as the only reference for identity. Prepare a horizontal sheet of eight large head and shoulder portraits in equal fields, four columns and two rows. In order from left to right: neutral, mildly joyful, angry, sad, surprised, worried, confident, determined. Change the expression of the eyes, eyebrows, and mouth, but keep the shape of the face, age, skin tone, short silver hairstyle, scar on the right eyebrow, mustard jacket, and navy scarf.
The same camera distance, soft light, and neutral matte background in each field. The whole head and both shoulders should fit in the frame. No signatures, foreign faces, new decorations, hairstyle changes, or skin beautification. If eight portraits become too small, create two boards of four emotions instead of reducing the faces.

Use the approved image of the full character as a reference for the silhouette and outfit, and the approved portrait as a reference for the face. Create a horizontal sheet of six full views of the same fictional heroine arranged in three columns and two rows. In order: neutral standing front; natural step; sitting on a simple stool; relaxed posture with weight on one leg; alert, tense posture; readiness to move without weapons or props.
Each panel shows the whole head, both hands, and whole shoes with extra space at the edges. Keep the same body structure, age, silver hairstyle, mustard jacket, navy scarf, dark pants, and brown shoes. The same scale of the character, neutral background, and diffused light. The pose and expression may change, but no character gets a different outfit or new items. No text and additional people.
How to check the result? Match the neutral expression with each emotion: the eyes, nose, jawline, and scar should remain recognizable. Then compare the standing character with the other poses: the length of sleeves, number of pockets, and layers of clothing should make sense even during movement. If two panels differ, correct that panel or generate it separately. Do not accept the entire sheet just because it looks impressive from a distance.
Example of use in film: in the first six-second shot, the heroine calmly reads a sketch at the table, in the second she hears something behind the door and looks up, in the third she stands up and takes a step. For the first shot, use the approved neutral portrait and the sitting view; for the second, the worried portrait; for the third, the standing view and step. In each prompt, describe the movement between states, and do not ask the model to replicate the entire grid of poses. Save the final frame of each accepted shot as a reference point for the next.
@Image 1 is the approved neutral portrait: solely identity, hairstyle, and scar. @Image 2 is the approved portrait with worry: a hint for the final reaction of the same person, not a new heroine. @Image 3 is the view of the sitting silhouette: mustard jacket, navy scarf, dark pants, and body proportions. Do not copy the gray background of the boards into the scene.
One continuous shot, 6 seconds, format 9:16. The heroine sits at a table in a small archive and for the first two seconds reads a sketch of a plant. She hears a quiet sound behind the door, slowly lifts her gaze, and turns her head towards it. At the end, she is slightly worried but still sitting; she does not stand in this shot. The camera is still, medium shot, with a window lighting her face from the left. The same face and outfit from the first to the last frame, natural change of expression between the two states. No cuts, new people, additional dialogue, or sudden lighting changes.
One full board: prompt to copy
If you need one file to show to the team, you can try to gather the material into a 4:5 board. Treat it as the first draft of the document: the generator may confuse a small profile, the side of the scar, or the text. Usually, it is better to prepare and approve separate images, and then compile the full board from them in the editor. However, if you still want to make the first attempt in one step, use this original template:
Attach one clear photo or illustration of the character. Create a vertical production board 4:5 dedicated solely to this one person. The source image is the primary reference for the face and visible parts of the outfit. Do not copy its background.
Arrange the board in clear fields: (1) main portrait and short profile with space for manually entering the name, role, and features; (2) full silhouette from the front, at a three-quarters angle, from the side, and from the back, in equal scale; (3) three close-ups of the head showing the geometry of the face, hair, and distinguishing mark; (4) four facial expressions: neutral, joyful, worried, and determined; (5) three simple poses: standing, walking, and sitting; (6) details of the outfit, shoes, and accessories; (7) samples of main colors and materials. Leave blank fields for signatures and color codes that we will fill in manually after checking the image.
In each field, there should be the same character: the same age, face shape, skin tone, hairstyle, proportions, outfit, and placement of distinguishing marks. If the back of the clothing is not shown in the reference, suggest the simplest logical completion and do not present it as a confirmed fact. No new person, alternative outfit, beautification, rejuvenation, style changes, additional props, or invented text. Neutral background, soft light, readable layout, and large panels of the face. Working name of the character: [ENTER NAME].
The most important limitation: an image from one shot does not contain real information about the back, exact height, or all profile details. The model infers them. If you need documentation of a specific person or costume, add real photos from several sides. Do not treat the generator as a 3D scanner.
How to turn a card into a video: practical process
The card is not a film or a 3D model. It helps select consistent references for every shot. In a video tool like Higgsfield, you can upload images and assign them roles; capabilities and file limits depend on the model. For example, Seedance (as of October 2026, version 2.5 is available on Higgsfield) accepts images and references for characters, products, and locations. The producer promises consistency, but no generator guarantees the same face, so evaluate the result frame by frame.
- Choose the shot and angle of the face
Start with five or six seconds and one simple action, e.g., the heroine approaches the table, stops, and looks at the sketch lying on it. If the film is to have a profile, include the approved profile; for a look at the camera, include the front portrait.
- Separate the roles of attachments
The portrait gives the face, the rotation gives the outfit and proportions, the image of the location gives the background. Do not ask the model to copy the neutral background of the board into the film. If you are using only one complex image, crop the appropriate panel so that the face does not occupy just a few pixels.
- Describe the action, camera, and fixed things
Write what the person does from beginning to end, what remains in the frame, and how the camera moves. Add duration, proportions, and direction of light. In the lock, list only the most important features that the model often changes.
- Do a test, correct one cause
Check the face in the initial frame, when the head turns, and in the last second. Compare the scar, hair, sleeves, and shoes. If the profile fails, change the profile reference; if the outfit, use a clearer view of the clothing. Do not change all parameters at once.
- Save the winning frames
When one shot looks good, export a representative frame and use it as the next reference or the first frame of the next scene, if the tool allows. This is often more effective than returning solely to the original large board.
@Image 1 is the approved portrait of the fictional heroine's face. @Image 2 is her body rotation: mustard jacket, navy scarf, dark pants, and shoes. @Image 3 is the interior of a small botanical archive, used solely as a location and source of light. Do not copy the studio background from @Image 1 or the layout of the board from @Image 2.
One continuous shot, 6 seconds, vertical 9:16. The heroine enters from the left, approaches a wooden table, stops, and looks at an open sketch of a plant. The camera is at chest height, medium shot, with very gentle hand movement. Soft light from the window on the left. Keep the same face, short silver hairstyle, scar on the right eyebrow, proportions, and full outfit. The right hand is empty, the left remains at the side. No new people, clothing changes, time jumps, cuts, or additional dialogue. The movement has a visible beginning, middle, and end.
If the source is an existing recording of a person, the source film should define gestures and timing. In that case, the card can help maintain appearance during editing, but it does not replace the person's consent to use their image. In scenes with a recognizable public figure, do not present the generated performance as a real recording or as their product endorsement.
One card or several: it depends on the task
A complete character board is useful for animation and recurring shots of a person. A product advertisement or a scene in multiple rooms requires different documents. There is no one mandatory "super sheet" for every project. It is better to select a small set of cards, each with one role:
| Type of card | What it keeps in check | When it is useful |
|---|---|---|
| Character and face | Features, hair, proportions, distinguishing marks. | The same person in many shots or episodes. |
| Outfit and props | Layers of clothing, accessories, what the person holds. | Changing the costume between scenes would be a mistake. |
| Product | Shape, color, materials, cap, print, and packaging. | Advertising a bottle, shoe, phone, or piece of furniture. |
| Location | Layout of doors, windows, furniture, direction of light. | Many shots in one room or outdoor setting. |
| Light and color | Light sources, contrast, material palette. | The scene should look the same after cutting. |
| Shots and movement | Frame scale, camera axes, gestures, cutting moments. | Film with editing, advertising, or action sequence. |
For a short film with one heroine, usually sufficient are: a portrait, view of the outfit, location reference, and a simple shot plan. For a perfume advertisement, the card for the bottle will be more important than the eight emotions of the model. For a small campaign, you can have separate PDF pages or several images in folders character/, product/, location/, and shots/. Name the versions and keep the approved set so that subsequent generations do not use random variants.
Example of another card: product
On the product board, we are interested in shape, materials, and characteristic details. The illustration below shows a fictional bottle from our guide on perfume advertising. It is also a good exercise in control: the two upper panels are too similar, although one was supposed to show a three-quarters angle. In practice, reorder the missing view or replace the panel with a real photo. The aesthetics of the board alone are not enough if the necessary information is missing.

Use the attached photo as the primary reference for one product. Prepare a horizontal board with four large fields: full front view, clear view at a three-quarters angle, true side profile, and detail of the most important material or mechanism. Keep the same shape, scale, colors, number of parts, and proportions. Studio lighting and neutral background identical in all panels. Do not copy the same angle twice; do not add a new version of the product. No signatures or invented logos. If the back or profile does not result from the photo, mark that view for later manual verification instead of presenting it as a certainty.
Example of another card: location
For a location, it is not about the face or style, but about the geography of the scene. In our gallery, the window is on the left, the pool lies in front of the stone pedestal, and warm light falls from the right. A board from different camera points helps to notice when the model adds a second window or moves the pedestal. The top view created by AI should be treated as a sketch, not an architectural projection.

Based on the attached photo, prepare a board of the same location from three realistic camera points: a wide shot at eye level, a diagonal shot after a slight shift to the right, and a low shot close to the floor. Keep the placement of windows, doors, pool, pedestal, and directions of light. Add a separate simple sketch of the layout of these elements from above. In each view, there is one room, not three variants. No new walls, additional windows, characters, or props. The top sketch is auxiliary and requires manual verification.
How to combine several cards in one prompt
Assign roles to the images before describing the action. Let one reference correspond to the face, another to the product, a third to the location; do not require the clothing photo to establish the identity of the model. In a film with several scenes, also include the approved frame from the previous shot if the tool supports a starting image. This helps connect separate generations. Example map:
@Image 1: solely the face and hair of the heroine. @Image 2: solely the outfit and proportions of the silhouette; do not take the face of the mannequin or person from the clothing photo. @Image 3: exact geometry and materials of the product; do not copy the studio background. @Image 4: layout of one room and direction of lighting. @Image 5: approved first frame of this specific scene. Maintain consistency with assigned roles; we describe the action, camera, and time below.
You do not have to attach all five files every time. Select only those that solve the problem of that shot. Too many similar but not identical images can introduce conflicting cues. If the interface allows for few references, choose the most important face or product and the appropriate first frame.
Identity lock in later prompts
For a new scene, include the approved portrait and the appropriate silhouette view. State the role of each image. For example: "@Image 1 is the face and hair; @Image 2 is the silhouette and outfit; do not copy the neutral background of the board." Only then describe the action and location. Do not add five nearly identical but slightly different portraits: conflicting references can weaken consistency.
The character in this scene is a fictional heroine from @Image 1 (face, hair, scar) and @Image 2 (silhouette, mustard jacket, navy scarf, dark pants). Keep her age, features, skin tone, proportions, and outfit. Do not rejuvenate, beautify, change hairstyle or accessories. Change only the pose, expression, and location described below.
Receiving the card: five questions
- Does the profile and three-quarters view still look like the same person?
- Are the scar, hairstyle, and clothing elements on the correct side?
- Do the hands, legs, and shoes have credible proportions?
- Are the signatures and color codes correct? If not, add them manually outside the generator.
- Do you have the right to use the person's photo and publish derivatives?
The card is a control tool, not a magical "freezing" of the character. In the film, check the face at the beginning, when the head turns, and at the end. For a quick start, we also prepared a minimalist card of three panels and a separate prop. Also, see our example of swapping two characters in one film, guide to references in AI film, and editing existing footage in Genjutsu.
Frequently Asked Questions
Is one photo enough for a character card?
For a simple shot, often yes. However, the model infers the back, profile, and exact height, so for an important project, add real photos from several sides and mark in the card what is inferred and what is fact.
Can I trust the color codes from the generated board?
No. Treat the text and HEX codes drawn by the generator as decoration. Manually enter the color using a color picker in the editor.
How many references should I attach to one shot?
As many as solve the problem of that shot: usually a portrait, view of the outfit, location, and possibly the first frame. Several similar but not identical portraits can give conflicting cues.
Can I make a card for a real person?
Technically yes, but you need that person's consent to use their image and publish derivatives. Do not present the generated performance as a real recording or endorsement.
Sources
| Source | Type | Status / credibility |
|---|---|---|
| Picsart AI Playground: list of available models | Manufacturer's page | Checked 23.10.2026; availability depends on the plan |
| Higgsfield: description of Seedance 2.5 and handling references | Manufacturer's page | Checked 23.10.2026; marketing description |
Status as of October 23, 2026. The material is for informational purposes and does not constitute legal advice. Models, limits, and prices are determined by service providers. Before publication, ensure you have rights to the photos and images used.
Comments (0)
No comments yet. Be the first!