How to illustrate a children's picture book with AI and keep the characters consistent

Build a consistent AI picture book with character references, reusable page briefs, targeted edits and editable layouts. Follow a complete illustration-to-proof workflow.

A small fox character studies an open sketchbook of matching fox poses
Dreamina
Dreamina
Nov 6, 2026

Start in Dreamina by approving your main character before illustrating the story. Save a clear reference image, a short list of defining features and a separate style reference. Use those same materials for each page, change the action and setting deliberately, then compare the result with the original design.

For a complete picture book, add a page plan and an editable layout. A recognizable face is only part of continuity: clothing, proportions, supporting characters, recurring places and the sequence of events also need to agree. The workflow below takes you from a character reference to reviewed illustrations and a book file you can proof.

Table of content
  1. A repeatable workflow for the whole picture book
  2. Plan the pages, spreads and final format
  3. Define what makes each character recognizable
  4. Keep the style guide separate from the character guide
  5. Build the first reference and page in Dreamina
  6. Choose supporting tools by their role
  7. Write one reusable page brief
  8. Prove the method on three different scenes
  9. Illustrate the book in small approved groups
  10. Stage two characters before adding detail
  11. Return to the original references when the book drifts
  12. Repair a detail without approving a new character
  13. Review the pages together
  14. Add the words as editable text
  15. Prepare the files for their actual destination
  16. Keep the source artwork and production notes
  17. Frequently asked questions
  18. Start with one character and a small sequence

A repeatable workflow for the whole picture book

Develop the character in Dreamina, approve a small sequence, then expand the book using the same references. Work in this order:

    1
  1. Divide the story into pages or spreads, with a clear visual purpose for each.

  2. 2
  3. Approve a character guide containing the original design, important views and fixed details.

  4. 3
  5. Choose the illustration style and save an appropriate visual reference.

  6. 4
  7. Make a few different scenes to check that the method suits the story.

  8. 5
  9. Illustrate small groups of pages, correcting differences before continuing.

  10. 6
  11. Add the words in a layout tool and inspect the complete sequence.

  12. 7
  13. Prepare the required export and review a proof for the intended destination.

This gives you checkpoints where a decision can still be changed without rebuilding the entire book. Reference images improve direction, but every illustration still needs a visual review.

Plan the pages, spreads and final format

Divide the story into visual moments before requesting finished illustrations. Decide what the reader should discover in each image and what the page turn should reveal. An illustration can introduce something the text leaves unsaid, rather than simply repeating every sentence.

Make a simple planning table. This proposed gardening story shows the level of detail that helps:

Story moment
Characters and action
Continuity to retain
Composition and text space
Introduction
A small robot finds a seed packet
Approved robot design and gardening bench
Show the whole character; reserve a calm area for the introduction
A problem
The robot looks at an empty pot
Same packet, pot and tools
Closer view of the action; keep the important object visible
An attempt
The robot plants a seed
Same gloves, backpack and pot
Show the hands clearly without crowding the text
Resolution
The robot notices a new leaf
Explain the passage of time through the story
Give the final discovery enough space to read emotionally

Choose the intended book dimensions early, using the printer or digital destination's requirements. A single page and a double-page spread need different compositions. Keep faces, hands and essential story objects away from the binding area. There is no single page count or image ratio that suits every picture book.

If you already have rough drawings, they can guide the staging before you develop finished artwork. The guide to turning sketches into images explains that separate preparation step.

Define what makes each character recognizable

Create a character guide that separates fixed design details from things the story can change. Start with a small number of visible features a child can recognize across different poses: the silhouette, face, proportions, colors and one or two distinctive details.

Record in the guide
Example for a small robot
Shape and proportions
Rounded cream body, large eyes, short limbs
Face
One amber eye and one teal eye, with the approved arrangement shown in the reference
Distinctive detail
A short antenna with a red ball
Clothing or equipment
Dark-blue gloves and a mustard backpack
Expression range
Curious, worried and pleased, without redesigning the eye shapes
Story changes
A tool held in the hands, a new pose or an approved change of outfit

Approve one clean full-body design. Develop the front, side, back and any useful expressions from that design, checking that they agree. A generated turnaround can contain three subtly different characters; inspect the joins, clothing, markings and body shape before adopting it.

For a second character, make another guide and record the relative size of the pair. Keep a clear distinction between the original reference and an illustration that happens to look attractive. The newest image should not automatically become the character's new identity.

Keep the style guide separate from the character guide

The character reference says who appears; the style guide says how the book is drawn. Record the medium, line treatment, palette, texture, background detail and approach to lighting. “Soft watercolor with restrained pencil outlines and warm earth colors” is easier to maintain than a request for a beautiful storybook image.

Use a style reference you are authorized to use, and explain its role. Ask for its visual treatment while using the character sheet for identity. Otherwise, a style image containing another person or animal can introduce unwanted facial features or clothing.

Consistency does not require identical lighting on every page. A night scene can be darker than a morning scene while retaining the same drawing language. Decide which changes tell the story and which would make the pages look as though they belong to different books.

Build the first reference and page in Dreamina

Dreamina is a useful starting point for designing the character and developing one illustration at a time. Choose the image-generation or editing workflow and the model available for that task. Describe the character's visible features, proportions and clothing, then specify the visual medium and lighting. An original design with a few clear identifying details is easier to direct than one that depends on dozens of tiny markings.

Dreamina AI image generator page with its text-and-photo prompt area and model selector

Approve a clean full-body image before requesting more views. Use that image as the reference for a side view, back view or expression, then compare the result with the original. A sheet with three inconsistent versions is not a reliable guide simply because it fits on one canvas. Correct the views before moving to a story scene.

For each page, attach the approved reference, identify its role and describe the new action, expression, setting and composition. Keep the same character specification and chosen style. If a page is nearly right, request a narrow correction and inspect the whole result again. Available reference and editing controls depend on the selected model; a request to preserve a detail is an instruction to review, rather than a guarantee of exact preservation.

Keep the story text out of the illustration while developing the artwork. Add it as editable text during layout, with room for revision and translation. Save the approved reference and the accepted page as separate files so you can return to either one.

A robot reference carried into one garden scene

The two images below show a Dreamina character reference and one new garden scene made with a paid Basic account. The reference shows the robot from three directions. In the scene, it holds a seed box at a planting bench, with its contrasting eyes, red-tipped antenna, dark-blue gloves and mustard backpack still recognizable.

Three-view reference of a cream robot with contrasting eyes, blue gloves and a mustard backpack
The reference robot holding a seed box in a garden-bench scene

Use a pair like this to inspect the design cues that carry into a new setting. It is one scene, rather than evidence that an entire illustrated book will remain consistent. The next step is to try a different pose and a closer expression, then review those images beside the same original reference.

Choose supporting tools by their role

Use Dreamina for the initial character and page development, then choose another tool for a specific illustration or layout need. You do not need to use every application in the table.

Tool
Useful role in the book
Reference or editing approach
Main decision to retain
Dreamina
Character development and individual illustrations
Approved images, a scene brief and controlled revisions
Which details define the character
ChatGPT Images
Conversational image development and corrections
Attached references and targeted editing instructions
What changes and what stays fixed
Midjourney
Reference-led artwork in a chosen visual direction
V7 Omni or the V8.X Edit Model, with suitable style guidance
The version and reference roles
Leonardo AI
Combining references and creative direction
Current Omni image references or supported specialized guidance
How each reference should influence the result
Adobe Firefly
Visual treatment and artwork refinement
Reference-led image creation and editing
Style and composition versus exact identity
Canva
A straightforward multi-page layout
Imported artwork and editable text or design elements
Page hierarchy, cropping and readability
Adobe InDesign
Detailed page setup and print-layout work
A document built around the required production settings
Trim, margins, bleed and the editable source
Book Creator
A digital or classroom reading edition
Images, text and optional audio in a book experience
The reading destination and sharing settings

Pick one illustration route for the first small sequence. Switching generators midway through production can change the visual treatment as well as the character. If another tool supplies a needed capability, check a few representative pages before moving the whole project.

ChatGPT Images: develop and revise through conversation

ChatGPT Images is useful when you want to explain an illustration change in ordinary language and continue refining the result. Attach the approved character artwork, identify any separate style reference and describe the new page. Specify the details that should remain fixed instead of relying on “the same character” alone.

For a correction, provide the page you want to edit and describe one change: restore the original coat color, remove an unintended accessory or adjust the expression. Review the complete returned image before accepting it. Keep the original reference available rather than letting a sequence of slightly different outputs become the new design.

The OpenAI image-prompting documentation also illustrates a character introduced in one scene and reused in another. That is an API example of a reference-led method; its model names, quality settings and image sizes are not ChatGPT or Dreamina interface instructions. Choose the controls available in the product you are actually using.

OpenAI official Images in ChatGPT guide describing image creation and editing

Midjourney: use the reference system for your version

Check the model version before following a Midjourney reference tutorial. Omni Reference belongs to V7. On the web, upload an image through the Imagine bar and assign it to Omni Reference; pin it when reusing it across prompts. The Discord route uses an image URL with --oref. V7 accepts one Omni image, so a shared cast reference is a different input from separate character files.

For V8.X, Midjourney directs users to the Edit Model, which supports up to four reference images. Do not combine instructions for older Character Reference, V7 Omni and the newer Edit Model into one recipe.

In V7, --ow adjusts reference influence from 1 to 1,000, with 100 as the default. The guide generally recommends staying below 400 unless using very high stylization. Style and identity guidance can compete; increasing a weight is not an identity guarantee. Fine details may differ, and Omni has its own GPU-time and editing restrictions. Check those before committing the book to that route.

Midjourney official Edit Model guide showing its reference-image approach and version support

Leonardo AI: assign clear roles to the reference images

Leonardo is useful when you want to combine character, visual treatment and scene references with explicit instructions. Its current Omni workflow accepts up to six images: open Image Guidance, choose Image Reference, upload the useful references and explain how they should combine. The home workflow can select a model automatically; Image Creation provides more control.

For a book page, identify the character image, style image and any composition reference separately. Ask for the character's approved appearance in the new scene, rather than expecting every input to contribute equally.

Leonardo's help also describes specialized Character, Style and Content Reference workflows. Those controls and their older limits are not interchangeable with the newer Omni route. Select the method available for the model you are using, then inspect the result. A style resemblance or a similar composition does not by itself establish the same character.

Leonardo official Image Guidance guide explaining the current Omni reference workflow

Adobe Firefly: guide the visual treatment and refine the artwork

Firefly is a useful option for exploring an illustration treatment and refining assets within an Adobe workflow. Begin with authorized reference material and describe the subject, medium and composition. Keep the character's defining details explicit while changing the scene.

Distinguish the roles of the available reference controls. A style reference guides the appearance of the artwork; a composition reference guides its arrangement. Neither should be treated as a promise that every facial feature or small marking remains identical.

Choose a page that needs a specific refinement—such as a simpler background or clearer space for the text—and review the returned artwork against the character guide. Keep exact wording and final typography editable during layout. Available models and features have their own conditions, so confirm the chosen workflow before planning a complete book around it.

Adobe Firefly official image-to-image page illustrating a reference-led transformation

Write one reusable page brief

Keep the character and style instructions stable while changing the page action and composition. A brief should identify the reference roles, fixed details, new action, expression, setting and space needed for text.

Here is a proposed brief for a new robot illustration. Adapt it to your own approved design and actual page dimensions:

Reference roles: Image 1 defines the robot's appearance. Image 2, if supplied, defines only the illustration's texture, palette and lighting treatment. Use the character design from image 1.

Character: A small rounded cream robot with the same eye arrangement as the reference, a short red-tipped antenna, dark-blue gloves and a mustard backpack. Retain the approved face shape, body proportions and equipment. Do not add another accessory or another robot.

Style: Use the approved book's visual medium, surface treatment and palette. Keep the same approach to outlines and background detail.

This scene: The robot stands beside the gardening bench and carefully places one seed into the waiting pot. It looks down at the pot with a curious expression. Keep the approved bench, packet and tools recognizable where they are visible.

Composition: Show the hands and pot clearly. Reserve an uncluttered area for the story text and keep important details away from the planned binding and trim areas. Use the dimensions specified for this page or spread.

Changes and constraints: Change the pose and action to match this scene. Preserve the robot design. Create the illustration without story text or page numbers; those will be added during layout.

The prompt states what you want. Compare the output with the reference before accepting that those instructions were followed.

Prove the method on three different scenes

Before illustrating the whole book, try a small sequence that exposes different consistency problems. Useful choices are a clear introduction, a scene with a different pose and a close view showing emotion. Choose scenes from your actual story, including a difficult recurring situation if there is one.

Keep the approved character and style references constant. Compare the face, body shape, equipment and drawing treatment. A method that produces a good front-facing portrait may still struggle with a side view, hands holding a prop or two characters together.

Estimate the work needed to finish these pages: generation, revisions, manual corrections and layout. If the method needs too much repair, simplify the design or adjust the workflow before producing dozens of illustrations. This small sequence is a production decision, not a reason to assume every later page will work automatically.

Illustrate the book in small approved groups

Work through a few pages at a time, comparing each one with the approved references before continuing. Generate an illustration, inspect it, correct it and save the accepted version. Then review the group in story order.

Use the page plan to track weather, time of day, clothing and important props. If a seed packet is opened on one page, the next page should not casually return it to an unopened state. If the characters return to a familiar room, use a location reference as well as the character images.

Keep rough drafts separate from accepted illustrations. A simple record of the page, prompt, references, chosen model and required corrections makes a later revision easier. It also prevents a discarded version from slipping into the book layout.

Stage two characters before adding detail

Give each character its own identity and a clear place in the scene. Name the characters in the brief, explain which reference belongs to each and specify their relative size. Describe the action in visible terms: the smaller rabbit stands to the robot's left and looks at the seed packet in its hands.

Check the arrangement before adding elaborate scenery. Faces, clothing and distinctive colors can be mixed between characters. A handover can create a second object, while a hug can merge arms or clothing. Simplify the action or prepare a clearer starting composition when the relationship is difficult to read.

The number and type of reference inputs depend on the selected tool and version. If a route accepts only one identity image, use an approved shared reference where appropriate and describe the cast clearly. Do not assume that several separate images will be accepted merely because another generator supports them.

Return to the original references when the book drifts

Use the original character guide as the authority, even when a previous page supplies useful scene context. Repeatedly editing the latest output can gradually normalize a different eye shape, a shorter body or a new accessory.

When you spot drift, place the affected page beside the original. Identify the difference in concrete terms and return to the approved design for the correction. A previous illustration can explain the room or lighting, while the original character reference continues to define identity.

Keep the model and important settings stable where practical, and record deliberate changes. If you switch methods, review the existing pages alongside the new outputs before continuing. A consistent file history is more useful than assuming a character name will make the generator remember the design.

Repair a detail without approving a new character

Repair one identifiable problem at a time. If the robot's gloves have become green, supply the original reference and the page to edit, identify the gloves and request their approved dark-blue appearance. State that the face, antenna, backpack, pose and composition should remain unchanged.

Compare the returned image with both inputs. Did the gloves improve? Did the eyes switch colors, the backpack disappear or the expression change? An edit can solve one problem while creating another. Save an accepted version only after that second check.

If repeated edits keep damaging an otherwise good page, return to the last approved version. A carefully composited correction or manual illustration work may be more predictable for a small feature that must remain exact. Keep the layered working file so the repair can be revised without rebuilding the whole page.

Review the pages together

Compare the book as a sequence, not only as separate attractive images. Make a contact sheet of the illustrations to spot gradual changes, then inspect individual pages at full size for small defects.

Review dimension
Questions to ask
Identity
Are the face, markings, silhouette and proportions still recognizable?
Wardrobe and equipment
Are changes intentional? Are accessories missing or duplicated?
Relative scale
Does the smaller character remain smaller when the pair shares a scene?
Style
Do texture, line treatment and color choices belong to the same book?
Place and time
Do doors, furniture, weather and recurring locations agree with the story?
Story action
Can the reader understand what happened and what changed?
Layout
Are faces, action and text safe from cropping, binding and clutter?

Read the sequence without the words as well. If the intended emotion or action is unclear, more descriptive text may not solve the illustration problem. For a shorter exercise combining recurring characters and dialogue, see creating a consistent four-panel comic.

Add the words as editable text

Add the story text after the illustration is approved, using a separate editable text layer. This makes proofreading, translation and changes to line breaks much easier than repairing words embedded in a generated image.

Choose readable typography, consistent spacing and enough contrast against the artwork. Review the page at its actual reading size. Move the text or simplify its background if it competes with the character's face or the main action.

Read the story aloud while turning the pages. Check the rhythm, punctuation and relationship between the sentence and illustration. For another language, review how longer wording changes the layout. Keep an editable source so a text correction does not require regenerating the artwork.

Canva: assemble an approachable page layout

Canva is useful for arranging approved illustrations and story text in a consistent multi-page design. Set up the intended page dimensions, import the artwork and add the text as separate elements. Use a suitable template as a starting point, then adapt its fonts, colors and spacing to the book.

Repeat a few purposeful layout choices, such as the text style and margins, while allowing different compositions where the story needs them. Check whether placing an image into a frame crops an ear, antenna or important object. Review a complete spread rather than only the editing canvas.

Inspect the exported file against the destination's requirements. A finished-looking template does not establish that every printer will accept its dimensions, color settings or bleed. Keep the editable design for final copy and layout changes.

Canva official eBook-design page showing an editable book layout example

Adobe InDesign: start with the printer's document settings

InDesign is worth considering when the book needs more detailed page-layout and print-production control. Start with File > New > Document, choose the appropriate category such as Print, and set the dimensions, page count, margins, columns and bleed values before creating the document.

Use the actual production specifications rather than assuming a preset is correct for your book. Place the approved artwork and editable story text into the layout, then review facing pages and the space near the binding. Keep the working document and linked illustration files available for revisions or a production handoff.

A blank document is a valid starting point. Adobe Stock templates have separate access and licensing requirements; you do not need to begin by buying a template. If you already have a layout workflow that meets the destination's needs, another application is a production choice rather than a mandatory purchase.

Adobe InDesign official guide showing document setup, dimensions, margins and bleed

Book Creator: choose a digital reading version

Book Creator is a useful option when the destination is a digital or classroom reading experience. Its educational workflow combines text, images and optional audio or video, with reading and sharing features suited to that setting.

Import the approved illustrations, arrange the story text and decide whether narration adds something useful. Check the reading order, image descriptions, text size and audio against the intended readers. Review the actual book experience and sharing settings before making it available to others.

A digital edition has different production needs from a printed book. Do not treat an online reading version or a downloadable file as automatic approval for commercial printing. Choose the destination first, then prepare and inspect the appropriate edition.

Book Creator official features page introducing its educational book-creation tools

Prepare the files for their actual destination

Use the printer or publishing destination's specifications before deciding that a file is finished. For print, confirm the trim dimensions, bleed, safe margins, binding area, required color handling and accepted export format. Check whether the printer wants single pages or spreads. Prepare the cover using the actual cover and spine template for the chosen binding and page count.

Calculate image requirements from the size at which the artwork will be placed. For example, if the printer asks for 300 pixels per inch and an image will fill eight inches of a page, it needs 2,400 pixels across that placed dimension. Bleed and a wider spread can require more. Upscaling changes the pixel count, but it does not prove that small features will print clearly.

Inspect a physical proof when producing a printed edition. Check crop, binding loss, color, small text and how the book reads in your hands. For a digital edition, inspect the actual reading device or app, including text size, page order and any narration. Keep those reviews separate from judging whether the illustrations look consistent.

Keep the source artwork and production notes

Keep the approved references, illustration files and editable book layout together. Save the accepted page versions, original source assets, relevant prompts and the model or method used. Record planned changes, such as a new outfit or a later time of day, so they are not mistaken for errors.

Keep permission records for uploaded artwork, photographs, fonts and other source material. Before commercial distribution, check the applicable generator terms and the publisher or platform's current requirements, including any AI-content declaration. Access to a tool and permission to use every asset in a book are separate questions.

Frequently asked questions

These details help keep the process repeatable as the story becomes more complex.

Can one prompt illustrate the whole book?

Use a separate brief for each illustration or planned spread. A grid of images can help explore a story, but it does not replace individual page composition, editable text, continuity review and a finished book layout. Plan the sequence first, then approve the artwork in manageable groups.

Will repeating the same prompt keep the character identical?

Repeating a description can help, but it does not guarantee the same visual identity. Supply the approved reference, repeat the important design details and compare each result with the original. Correct differences before using the image as context for another page.

How many reference images should I use?

Start with a small set in which every image has a clear job. One may define the character, another the style and another the location or pose. Avoid conflicting designs, and check the selected tool and version's input limits. More references are not automatically more useful.

Must the character wear the same outfit on every page?

The outfit can change when the story calls for it. Create an approved wardrobe variant and record where the change happens. Keep the face, proportions and other identifying details recognizable, and check that the clothing does not change accidentally before or after that story moment.

What resolution should the illustrations have?

Choose resolution from the intended display or printed size, rather than a label such as HD or 4K. Use the destination's specifications and the artwork's actual placed dimensions, then inspect the final export or proof. A high pixel count alone does not establish clear small details or a suitable book layout.

Do I need to train a custom model?

Start with approved references and a short scene sequence before committing to custom training. A more technical workflow may suit repeated production, but it adds data preparation, permissions, configuration and maintenance. Compare that effort with the corrections your reference-led workflow actually needs.

Start with one character and a small sequence

Create and approve the first character in Dreamina, then develop a few deliberately different scenes. Keep the original design, style reference and page plan available throughout production. Once that small sequence works, expand the book, add editable text and review the complete layout for its intended readers.

Hot and trending

98% OFF—Limited time

98% OFF—Limited time

98% OFF—Get started for just $0.29/month

Try now