To keep the same character across a comic or picture book, write one fixed description, make one clear reference picture from it, then attach that picture and repeat the description in every scene. Check each new page against the reference before you move on.
This is the method OpenAI’s own image guide gives for a children’s book, and Google’s image guide has a comic-panel template that works the same way.
- Write the character down first. A short, fixed description, copied word for word into every prompt.
- Make one reference picture, plain and front-on, and reuse it, not the latest drifted page.
- Keep the design simple. Fine details such as freckles or logos may not match, Midjourney warns.
- Check every page against the reference, and add speech-bubble text outside the AI tool.
- Publishing has rules: Amazon KDP asks you to tell it about AI-generated images.
An AI image model does not remember the hero of your story from one picture to the next. Think of a casting call where a new actor turns up for every scene, and you have only a costume list and a photo to make them look alike. The better the list and the photo, the closer the actors look.
How do you keep a character the same across a story? Six steps
The plain point: the work is mostly done before the first page. The table is this site’s routine, built on the makers’ guides quoted below.
| Step | What you do | Why it helps |
|---|---|---|
| 1. Write the character sheet | Do: name, shape, face, hair, one outfit, colours, one signature item, art style. | Because: the same words in every prompt give the model less room to wander. |
| 2. Make the first picture | Do: one plain, front-facing picture on a simple background. | Because: it becomes the reference every page is checked against. |
| 3. Add other angles | Do: side, back and three-quarter views, made from the first picture. | Because: the tool then has more than one view of the character. |
| 4. Attach the reference | Do: upload the picture with every scene prompt, where the tool allows. | Because: the picture carries what words cannot. |
| 5. Generate one scene at a time | Do: change the place and action; repeat the character words. | Because: only the scene should change. |
| 6. Check, then keep or redo | Do: lay the new page beside the reference. | Because: small drift is easy to fix early and hard to fix at page 20. |
What goes in a character description?
The plain point: the things a reader would notice at a glance, in plain words, and nothing you do not need.
OpenAI’s own example for a children’s book defines the character’s “appearance, proportions, outfit, and tone” and adds the constraint “Original character (no copyrighted characters)”.1 Here is a template in that spirit. The character is invented for this page.
Keep the sheet short and visual. Midjourney warns that “intricate details like specific freckles or logos on clothing may not perfectly match your reference”, so a character with a big, simple signature item, such as the scarf above, is easier to hold than one covered in small marks.2 (The advice about a big, simple item is this site’s reading of that warning.)
How do you make the first picture and the other angles?
The plain point: make one clean picture, then build the other views from it, not from scratch.
Runway recommends reference images with “Natural, even lighting”, “Moderate quality” and “Neutral subject expression”, because they give “a ‘blank canvas’ that simplifies transformation”.3 Google’s guide describes the next step: “You can generate 360-degree views of a character by iteratively prompting for different angles. For best results, include previously generated images in subsequent prompts to maintain consistency.”4
Make a front, a side, a back and a three-quarter view, and pick the best of each. This is the “character sheet” you will attach later. For a longer turnaround prompt, see character consistency in AI.
How do you keep the character the same in ChatGPT, Gemini, Midjourney and Runway?
The plain point: all four let you attach your character picture. They differ in how many pictures and what they call it.
| Tool | What the maker says to do | How many pictures |
|---|---|---|
| ChatGPT / OpenAI | Do: reuse the generated character image and describe a new scene, repeating the appearance constraints. | Number: not stated in the guide read. |
| Gemini (Nano Banana) | Do: mix reference images; the guide has a comic-panel template. | Number: up to 14 in all, of which up to 4 or 5 can be characters, by model. |
| Midjourney | Do: the Edit model with reference images (V8.1, V8.2); Omni Reference on V7. | Number: up to 4 with the Edit model; one with Omni Reference. |
| Runway | Do: tag a Reference and call it by name with @. | Number: up to 3 per generation. |
OpenAI’s steps for the next page: “Reuse each model’s generated character image and describe a new scene. Repeat the appearance constraints so the character stays consistent.”1 Here is a scene prompt in that style:
Google says its Gemini image models accept “up to 14 reference images”, with “Up to 4 images of characters to maintain character consistency” on Nano Banana 2.1, and its guide includes a comic template: “Make a 3 panel comic in a [style]. Put the character in a [type of scene].”4 Midjourney says its Edit model can “Generate new images using up to 4 reference images (replacing Omni Reference and Character Reference)”, and matches the aspect ratio of the first image you upload unless you add --ar.5 Runway says to “Use up to three References for a single generation” and that for complex changes you should “iterate elements individually, and use the resulting images as new references”.3
Whatever the tool, attach the same character picture and repeat the same character words on every page. Change only the scene.
What if two characters share a page?
The plain point: give each character their own picture and their own name, and say who stands where.
Midjourney’s tip: “If certain features or attributes are getting mixed up between your characters, it can be helpful to combine your characters into a single reference image.”5 This site’s page on identity lock prompt templates has a two-character template.
What do you do when the character starts to change?
The plain point: go back to the reference picture, not the last page.
OpenAI warns that “Repeated edits can still change details you intended to preserve.”6 So if page 12 looks off, do not use page 12 as the reference for page 13: start again from your original character sheet. This is this site’s advice. Compare pages side by side at full size, and watch the small things first: ear shape, eye spacing, the exact shade of the scarf.
What about words and speech bubbles?
The plain point: check every letter, and consider adding the words yourself.
Google’s comic template comes with a caution: “For accuracy with text and storytelling ability, these prompts work best with Gemini 3 Pro and Gemini 3.1 Flash Image.”4 In OpenAI’s children’s book example the prompts say “No text”. In this site’s use, leaving the pictures blank and typing the words in a layout or design program makes a misspelt bubble much less likely to reach the page.
Can you publish a book made this way?
The plain point: often yes, but you may have to say AI made the pictures, and you may own less of them than you expect.
Amazon KDP says: “We require you to inform us of AI-generated content (text, images, or translations) when you publish a new book or make edits to and republish an existing book through KDP. AI-generated images include cover and interior images and artwork.” It adds that using an AI tool to create the content counts as AI-generated “even if you applied substantial edits afterwards.”7 Other publishers and stores set their own rules, so read those of the one you use. On ownership, see who owns what you generate: the US Copyright Office concludes that “prompts alone do not provide sufficient human control to make users of an AI system the authors of the output.”8 Midjourney also notes that you “must have the necessary rights to use the images you upload”, so do not use someone else’s artwork or photos as your reference.2
Questions people ask
Can ChatGPT keep the same character across a whole book?
OpenAI’s guide describes exactly this workflow, one character image reused with the appearance written out again for each scene, but it does not promise a perfect match. Check every page against your first one.
Do I need to train a model for a consistent character?
Not to start. Reference pictures and a fixed description are enough for many projects. If you need the character across hundreds of images, or a tool that takes no references, training a small add-on is the next step: see how to train a character LoRA.
Can I use a photo of a real person as my character?
Only with their permission, and some tools refuse. This is this site’s advice; voice and likeness law explains why. For a story, an invented character avoids the question.
Why does my character look different on every page?
Each picture is generated fresh, so any change in the words or the reference gives the model room to drift. Repeat the same character words, attach the same reference, and start each page from the original sheet. Character consistency in AI goes through the causes.
Where to go next
- “Give me fill-in templates.” See identity lock prompt templates.
- “What are all the methods?” Read character consistency in AI.
- “How do reference images work?” See how to use reference images.
- “I need to change just one thing on a page.” See image edit prompts that change one thing.
- “I want to train my own add-on.” See how to train a character LoRA.
- “Who owns the pictures?” See who owns what you generate.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What should you write down before drawing the first page?
- A short, fixed description of the character to repeat in every prompt
- A different description for each page
- Only the name of the story
Show the answer
Answer: A short, fixed description of the character to repeat in every prompt. The page says to write the character down once and copy the same words into every prompt, so the model has less room to wander.
2. If page 12 looks off, what should page 13 start from?
- Page 12, since it is the latest
- The original character sheet
- A brand new description
Show the answer
Answer: The original character sheet. The page says repeated edits can change details, so go back to the reference picture, not the drifted page.
3. Why does the page suggest adding speech-bubble words outside the AI tool?
- AI tools cannot make any pictures with text
- It makes a misspelt bubble much less likely to reach the page
- Publishers require it
Show the answer
Answer: It makes a misspelt bubble much less likely to reach the page. The page says text in AI pictures needs checking letter by letter, and typing the words in a layout program keeps errors out.
- 1OpenAI, Image prompting, read at source 27 Oct 2026“Keep a character consistent. For a book with multiple illustrations, create a reusable character reference to help preserve appearance across scenes, poses, and pages. Change the environment and story while repeating the character’s defining details. Establish the character: Define the character’s appearance, proportions, outfit, and tone.” Its sample prompt includes “Constraints: - Original character (no copyrighted characters)”. ↑a“Continue the story. Reuse each model’s generated character image and describe a new scene. Repeat the appearance constraints so the character stays consistent.” Its sample constraints include “Do not redesign the character”. ↑b
- 2Midjourney, Omni Reference, read at source 27 Oct 2026“Account for Details: Be aware that intricate details like specific freckles or logos on clothing may not perfectly match your reference.” ↑a“You are responsible for all content you input, create, and share. You must have the necessary rights to use the images you upload.” ↑b
- 3Runway, Creating with Gen-4 Image References, read at source 27 Oct 2026“For optimal results with fewer iterations, use character images that meet these criteria: Natural, even lighting; Moderate quality; Neutral subject expression. These recommendations provide a ‘blank canvas’ that simplifies transformation”. ↑a“Article highlights: Tag your References to save them for future use; Use up to three References for a single generation; … For complex changes with more control, iterate elements individually, and use the resulting images as new references.” ↑b
- 4Google, Gemini API: image generation, read at source 27 Oct 2026“You can generate 360-degree views of a character by iteratively prompting for different angles. For best results, include previously generated images in subsequent prompts to maintain consistency. For complex poses, include a reference image of the selected pose.” ↑a“Nano Banana models allow you to mix up to 14 reference images.”; its table lists “Up to 4 images of characters to maintain character consistency” and “Up to 5 images of characters to maintain character consistency” in two of its model columns, and the comic template: “Make a 3 panel comic in a [style]. Put the character in a [type of scene].” ↑b“6. Sequential art (comic panel / storyboard) Builds on character consistency and scene description to create panels for visual storytelling. For accuracy with text and storytelling ability, these prompts work best with Gemini 3 Pro and Gemini 3.1 Flash Image.” ↑c
- 5Midjourney, Edit Model, read at source 27 Oct 2026“Generate new images using up to 4 reference images (replacing Omni Reference and Character Reference).”; “The Edit model will automatically try to match the aspect ratio of the first image you upload. To change this, simply include your desired aspect ratio in the prompt using the --ar parameter.” ↑a“Pro tip: If certain features or attributes are getting mixed up between your characters, it can be helpful to combine your characters into a single reference image.” ↑b
- 6OpenAI, Image prompting, read at source 13 Oct 2026 (re-used from this site’s upscaling page): “Repeated edits can still change details you intended to preserve. Restate those constraints and inspect each result.”↑
- 7Amazon KDP, Content Guidelines, read at source 27 Oct 2026“We require you to inform us of AI-generated content (text, images, or translations) when you publish a new book or make edits to and republish an existing book through KDP. AI-generated images include cover and interior images and artwork.” “If you used an AI-based tool to create the actual content (whether text, images, or translations), it is considered ‘AI-generated,’ even if you applied substantial edits afterwards.” ↑
- 8US Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, January 2025, read at source 13 Oct 2026 (re-used from this site’s common AI misconceptions page): “The Office concludes that, given current generally available technology, prompts alone do not provide sufficient human control to make users of an AI system the authors of the output.”↑
Read at source 27 Oct 2026: OpenAI, Image prompting; Google, Gemini API image generation; Midjourney, Edit Model and Omni Reference; Runway, Creating with Gen-4 Image References; Amazon KDP, Content Guidelines. Re-used from this site’s common AI misconceptions page (read 13 Oct 2026): US Copyright Office, Part 2: Copyrightability.
The six-step routine, the character prompts (Pip is invented for this page), the advice on drift and on lettering, and the advice on real people are this site’s own, not claims from these sources. No tool on this page was tested for this guide. Tool limits change, so check each maker’s page before you rely on a number.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK