How to Keep Characters Consistent in Illustrations

A common question: "I generated an image of my character in chapter one. How do I make her look the same in chapter five?" The answer on Magic Chapters is a Character Consistency image model. Instead of describing the character in words every time, it takes the character's portrait as a reference image and keeps the same face, hair, skin tone and body in every illustration you generate.

This guide shows you how to switch to a Character Consistency model, how to give each character a portrait, which model to pick, and what reference images can and cannot do. Generating images is a Premium feature, and every Character Consistency model is a Premium model.

How it works

Every illustration starts with an image description written from the scene. A Creative model turns that description into a picture from scratch, so the same character can get a different face in every image. A Character Consistency model gets the description and the portraits of the characters in the scene. It is told to keep each character's facial features, hair, skin tone and body exactly as in the portrait, and to take the pose, expression, clothing, setting and lighting from the scene.

Three things follow from this:

• The portrait is the identity. A character without a portrait is described in words, like with a Creative model. If none of the selected characters has a portrait, the image is generated by the model's Creative counterpart instead.
• The setting is set once. The model is chosen in your account settings and applies to every image in every story: chapter covers, scene illustrations made with Visualise and regenerated covers.
• Up to 3 portraits per image. Only the first three selected characters that have a portrait are sent as references.

Step 1: switch to a Character Consistency model

1. Go to Content Preferences → Images.
2. Under Image Generation Model, click the Character Consistency tab. On a free account this tab takes you to the Premium page instead.
3. Under Model, click the model you want. Each card shows an estimated generation time. If you are unsure, see the model recommendations below.
4. Under Image Style, keep or change your art style. The portrait and the illustrations should use the same style, so decide on the style before you generate portraits.
5. Click Update.

If you set up Premium with the Bring it to life step of the welcome wizard and chose Character Consistency there, this is already done. The wizard picks the recommended model for you, and you can change it on the same settings page at any time.

Step 2: give each character a portrait

The portrait is the image shown on the character's card in your library. With a Character Consistency model it is also the reference image for every illustration. Characters you create with a filled-in appearance get a portrait automatically, and so do characters extracted from a chapter (see How Character and Location Extraction Works). To check or replace it:

1. Open Characters, open the character and go to the Portrait section of the edit page.
2. Click Change. The Change Image window shows your image library. Pick an image and click Assign, or click Generate new to create a fresh portrait.
3. When you generate, the description is pre-written from the character's Appearance fields: Ethnicity, Skin Tone, Eye Color, Hair Style, Hair Color, Body Type and Appearance Details. Edit the text if you like, click Generate, then assign the result.
4. Repeat for the other main characters of the story.

What makes a good reference portrait:

• A clear, front-facing headshot. Face, hair and shoulders, in good light, with nothing covering the face. Generated portraits are framed this way on purpose.
• A neutral expression. The expression in the portrait is not copied; the scene decides whether the character smiles or frowns. A neutral face gives the model the cleanest starting point.
• The same art style as your illustrations. A photorealistic portrait used with an anime style, or the other way round, gives mixed results. If you change your Image Style later, generate new portraits.
• One person only. A portrait with two people in it confuses the model about which one is the character.

If a character looks different in some stories, for example a younger self or a disguise, give each look its own version with its own portrait. Only the version you pick for the chapter is used as a reference. See How to Use Character Versions.

Step 3: generate the illustration

Nothing changes in how you illustrate. The only thing to watch is which characters are selected, because only selected characters are sent as references:

• Scene illustrations: open the chapter in Edit mode, select a passage and click Visualise. Under Characters in image, make sure the right characters are ticked before you click Generate image.
• Chapter covers: move the mouse over the cover and click Regenerate image. Under Who's in the picture?, select the characters that should appear. With nobody selected, the characters are chosen from the story.
• New chapters: the automatic cover uses the characters attached to the chapter in the story form.

The illustrations tutorial covers the rest: placing images in the text, editing them and exporting the illustrated story.

Which model to choose

All Character Consistency models do the same job, but they differ in speed, in how well they keep a likeness, and in what they are allowed to show. The card of each model shows its estimated generation time.

• Seedream 5.0 Flash Edit is the recommended starting point and the model the Premium welcome wizard picks. It is fast and keeps faces well.
• Seedream V5 Lite Edit and Seedream V4.5 Edit are the earlier versions of the same family. Try them if a face drifts with the Flash model; they take about twice as long.
• Wan V2.7 Edit and Wan V2.6 are strong alternatives with a slightly different look and a similar speed. Good for a second opinion when the Seedream models keep making the same mistake.
• Flux 2 Turbo Edit is the fastest model of the group, and together with Qwen Image 2.1 Edit it is good at stylised art and clean compositions.
• Muse Edit is the strictest model and rejects any suggestive content. Choose it when your stories are fully family-friendly and you want the fewest moderation surprises.

Switching models costs nothing: every generated image counts the same toward your daily limit, whichever model made it. Try the recommended model with one scene, and only move on if a face keeps drifting.

Limits of character consistency

• Best with one or two characters. With a single character the likeness is usually very close. With two it is still good. Three works but is less reliable: features can be mixed between people, or one character can be dropped. More than three is not supported; the extra characters are described in words only.
• The scene wins for everything except the face. Clothing, pose, expression, hairstyle details such as "hair tied up", lighting and setting all come from the scene description. If the character must wear a specific outfit in every image, put it in the description, not in the portrait.
• The portrait does the heavy lifting. A blurry or badly lit portrait gives a vague likeness. If a character never looks right, generate a new portrait before you blame the model.
• Small drifts still happen. Eye colour, a scar or a tattoo can go missing in a wide shot. Mention the detail in the description, or fix it afterwards with Edit image so the rest of the picture stays the same.
• Styles need matching portraits. After changing your Image Style, regenerate the portraits so the reference matches the new look.
• Not available in the standalone image generator. The Generate new tab of your image library uses Creative models only. Character Consistency applies to illustrations made from a story.
• Premium only. All Character Consistency models need a Premium account.

Why you cannot upload your own photo

There is no way to upload a photograph to Magic Chapters, for a character or for anything else. Every portrait is generated by our own tools from a written description, and only those generated portraits are used as reference images. This is a deliberate choice for compliance and safety: a platform that generates intimate content and accepts real photos could be used to make deepfake-style images of real people, and we do not want to be part of that. It also means no photograph of anyone is ever sent to an AI provider on our side.

What you can do instead is describe the look you have in mind in as much detail as you want. Fill in the Appearance fields and write freely in Appearance Details: face shape, freckles, the gap between the teeth, the way the hair falls, glasses, a favourite piece of jewellery. Generate a few portraits, keep the one that matches best, and that look is locked in for every illustration from then on.

Final tip

Start with one character. Give her a clean, front-facing portrait in the art style you plan to keep, switch to Seedream 5.0 Flash Edit, then open a chapter in Edit mode and Visualise two different scenes with only that character ticked. If the two faces match, add the second main character and repeat. Build up to three only when the scene really needs all of them in one frame.