From Idea to Illustrated Book With AI, Without Losing the Story

AI has made illustrated books dramatically cheaper to produce and no easier to write. The images arrive in days; the story still takes as long as it always did. This is the order of work I use, and the reasons behind the order.

1. Premise and reader before anything else

I write two sentences: who the book is for, and what they will feel at the end. A book for a seven-year-old being read to at bedtime is a different object from a book a fourteen-year-old reads alone — different sentence length, different page count, different amount of white space. Choosing the reader early makes a hundred later decisions obvious.

2. Manuscript first, always

Generating art before the text is written is the most common and most expensive mistake in AI publishing. Illustrations made for a story that later changes are wasted, and worse, they start dictating the plot: writers begin bending scenes to fit pictures they like. I finish a full draft, read it aloud, cut roughly a fifth of it, and only then think about images.

Reading aloud is not optional for illustrated books. Rhythm is the difference between a page a parent enjoys on the tenth reading and one they start skipping.

3. Art direction as a written document

Before generating, I write the visual bible: the palette, the medium the art should imitate, line quality, level of detail, how characters are proportioned, and how light behaves in this world. Then character sheets for the cast, exactly as in comic work — identity blocks with fixed wording, reference views, and a canonical outfit.

The single sentence that saves the most rework: decide the style once, write it down, and reuse the exact wording in every prompt.

4. Shot-list the book

I break the manuscript into spreads and decide what each illustration must do: establish a place, carry an emotion, reveal information the text withholds, or give the eye a rest. Illustrations that merely repeat the sentence beneath them are dead weight. Varying scale across spreads — full bleed, vignette, small spot — gives the book a pulse.

5. Generate, curate, repair

Batch by character and by scene, keep alternates, and expect to hand-repair a meaningful share of the images: eyes, hands, continuity of clothing, background objects that drift between spreads. A final colour grade across every image makes work generated across different weeks look like one artist's hand.

Text inside illustrations is set in layout, never generated. Signage, book covers within the story, letters a character reads — all of it typeset by hand so it is spelled correctly and can be edited later.

6. Layout and typography

Generous margins, a readable serif or humanist sans at a size appropriate to the reader's age, and text placed in calm areas of the artwork rather than fighting it. Text over busy illustration needs a soft plate behind it or a deliberately quiet zone generated for that purpose. Page turns are structural: end a spread on tension, resolve it after the turn.

7. Cover last

The cover is a marketing object, not the prettiest interior picture. It has to work as a thumbnail: one clear focal idea, high contrast, title legible at the size of a postage stamp. I design three and test them small, on a phone, next to competing titles.

8. Production checks before release

• Full proofread by someone who has not read the draft before.
• Character continuity checked spread by spread against the sheets.
• Images at print resolution with bleed where a printer requires it; digital editions checked on a phone, a tablet and an e-reader.
• Front matter complete: title, copyright, credits, and a clear note disclosing AI-assisted illustration.
• Metadata written deliberately — title, subtitle, description and categories are how readers find the book at all.

Disclose, and price for the work

I state plainly in every book that illustrations are AI-assisted under human art direction. Readers respect the honesty, and it avoids the far worse outcome of being found out later. As for price: an illustrated book that took six weeks of writing, direction, repair and layout is not a cheap product because part of it was generated. You are selling the finished thing, not the method.

The tools removed the barrier that used to stop most writers — the cost of illustration. What they did not remove is the part that always mattered: a story worth turning the page for.