AI children's book illustrations: same character, every page

To make children's book illustrations with AI, settle the main character in one approved picture first, then generate every page with that picture as a reference image, the character's description repeated word for word, and the page's scene. Make each picture the shape of the printed page, including bleed, leave calm space where the words will go, and set the text yourself in a layout tool.

Print numbers come from Amazon KDP's Help pages Set Trim Size, Bleed, and Margins, Paperback Submission Guidelines, and Content Guidelines. Sume facts come from the Image API docs and the Sume API reference. All were read on 2026-09-28.

How do I illustrate a children's book with AI?

• Write a page plan: for each page, the scene, who is in it, and where the text sits.
• Design the main character in your book's style, pick one clear full-length picture, and host it at a public HTTPS URL of your own. Reference URLs must be public HTTPS.
• Write one style line and one character description, and paste both unchanged into every page's prompt. Sume's image API serves no seed, so the reference picture is what carries the character; consistent character AI image generator explains the technique.
• Generate each page on ChatGPT Image 2.5 with the character picture in input_references and the page size in image_size. It takes up to 16 references, enough for a small cast and a recurring room.
• Check every page against the character picture, and regenerate pages where the face, hair, or clothes drift.
• Lay out the pages and text in a design tool. KDP requires a PDF manuscript when images bleed to the edges of the pages.

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "Picture book illustration, soft watercolor and colored pencil. Mia: a six-year-old girl with curly black hair in two puffs, a yellow raincoat and red boots. Scene: Mia jumps into a big puddle in a city park. Keep the top third plain pale sky for text. No words or letters.",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/mia.png" } }
    ],
    "image_size": { "width": 2864, "height": 2896 },
    "n": 2
  }'

What size should each illustration be?

The printed page plus bleed. KDP trims 0.125 inch from the top, bottom, and outside edges, so art that runs to the edge needs a page 0.125 inch wider and 0.25 inch taller than the trim size, and KDP asks for images of at least 300 DPI. ChatGPT Image 2.5 takes an exact custom image_size: both edges in multiples of 16, at most 3840 on the long edge, and 655,360 to 8,294,400 pixels. The table shows the largest size within 1% of each page's shape.

Only the 8.5 × 11 in trim falls short of 300 DPI at these sizes. Enlarge its pages with Image Upscale 1.0, which takes an upscale_factor from 1 to 4, and check the returned size; AI image for print walks through the pixel math. A scene across two facing pages is at least 16 inches wide for these trims, but 3840 pixels, the longest edge ChatGPT Image 2.5 takes, cover only 12.8 inches at 300 DPI. So generate the spread as one wide picture, upscale it, and split it into two page files: KDP requires single-page files, not spreads.

How do I leave room for the text?

Say where the words go in every page's prompt, such as "keep the top third plain pale sky" or "leave a quiet area at the bottom", and add "no words or letters" so the model doesn't draw its own. Set the text in your layout tool, where it stays editable. KDP's minimum font size for interiors is 7 points, and in a book with bleed the top, bottom, and outside margins must be at least 0.375 inch, so keep the words that far from those edges.

Do I have to tell KDP the pictures are AI-generated?

Yes, if you publish on KDP. Its content guidelines say: "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." They add that "AI-generated images include cover and interior images and artwork." Pictures you created yourself and only refined with AI tools count as AI-assisted, which KDP does not ask you to disclose. Other publishers and platforms set their own rules.

What are the limits?

• No setting guarantees the same face on every page, so check each one.
• Each version is billed, and a failed or cancelled generation is not. ChatGPT Image 2.5 estimates each request's price from the size, the quality, and the images and text you send, and usage.cost in the response is the USD amount billed.
• High quality, this model's default, is one of the slow configurations most likely to return 202 with a job to poll instead of the images.
• Result URLs are Sume-hosted and signed, so download every page you keep.
• The cover is separate art with its own sizes; see AI book cover generator.