Flip through a good picture book and the pages feel like one place, even when the characters change. That’s no accident. The illustrator worked out the world’s rules before drawing anything: the weather, the materials, which colours appear and how often.
AI image tools can do the same, but only if you give them those rules. Prompt each page from scratch and you’ll get ten lovely pictures from ten different books. This walkthrough builds one small world instead, an old apple orchard on a misty morning, using CapCut’s AI image generator for the examples. The method works in most modern image tools.
Write the world before the characters
Start with a place you can describe precisely. “An enchanted countryside” leaves the model too many choices. “An old apple orchard on a damp, overcast morning” settles the trees, the ground, the weather and the light all at once.
Those details carry from page to page even when the main character changes.
Give colours specific jobs, too. In this orchard, sage greens fill most of the frame. Russet clothing picks out a person. A faded ochre ribbon marks one branch. When every colour shows up everywhere at equal strength, nothing stands out and the pages blur together.
Before any scene, write a short world description and keep it handy:
“An old apple orchard in early autumn after rain. Gnarled trees, damp grass, silver-gray bark, weathered timber, and soft morning mist. Cool diffuse daylight, natural photographic detail, muted sage greens and browns, small russet and ochre accents. A quiet, curious mood.”
It helps to turn that description into a one-page “world bible” you reuse for every prompt:
| Element | Rule for this book |
| Place | Old apple orchard, gnarled trees, timber gates and rails |
| Weather | Early autumn, just after rain, low mist |
| Light | Cool, diffuse, overcast. No hard sun and no golden rim light |
| Base palette | Sage green, brown, silver-gray |
| Accents | Russet (people), ochre (the ribbon) |
| Textures | Wet grass, weathered wood, wool, mud |
| Recurring clue | One faded ochre ribbon |
| Mood | Quiet, curious |
Paste the relevant lines into every prompt. That repetition is what makes the pages feel connected.
Page one: a curious child
The first scene introduces the orchard through one small detail. A girl has stopped to look at a ribbon tied to a branch. Her stillness makes the ribbon matter, without explaining it yet.
Put her to one side so the orchard stays visible around her. A tight portrait would tell us about her face. A wider frame tells us about the world she’s walking into, and on page one the world is the point.
Starting prompt:
“One horizontal, photorealistic storybook scene in an old apple orchard. A fictional girl with a long dark braid, russet knitted cardigan, slate-blue trousers, and muddy brown boots stands on the right, looking at a faded ochre ribbon on a low branch. Her arms rest naturally at her sides. A wet grass path passes between gnarled apple trees into morning mist. Silver-gray bark, a few muted red apples, cool diffuse daylight, natural skin and wool texture. Keep the left side quiet, with distant trunks and soft mist. No lettering, border, or additional people.”
One subtle point: a warm-coloured cardigan isn’t warm light. The wool can be russet while the light across the scene stays cool. Models often mix the two up and wash the whole frame in gold. If that happens, say so explicitly in a follow-up prompt.
Page two: a caretaker who belongs there
A second person tests the world far better than another picture of the first one. The caretaker needs his own silhouette, but nothing about him should suggest a different season or a different book.
Weathered clothes tie him to the orchard’s old timber. A muted russet scarf nods to the girl’s cardigan without dressing them as a matching pair. The ochre ribbon appears again as a quiet point of recognition.
This time, use your best image from page one as the visual reference, then describe a new view:
“Create a different horizontal photographic scene in the same orchard world. One fictional elderly caretaker with silver hair, a short gray beard, a moss-gray wool jacket, muted russet scarf, dark trousers, and muddy boots stands beside a low weathered timber gate on the left. He looks along the path with a patient expression, hands relaxed. An ochre ribbon hangs from a nearby apple branch. Keep the first scene’s early-autumn foliage, damp grass, gray bark, soft mist, and cool overcast daylight. Leave the right side quieter for future story text. No child, lettering, or border.”
The reference image carries the setting. The text carries the new cast member and the framing. Don’t copy the girl’s pose or put the gate in the same spot just to force a match. Readers should recognise the place while discovering a new corner of it.
That split between a fixed anchor image and the variable details layered on top is the core of most consistency workflows. It works much the same way when you keep AI characters consistent across a larger batch.
Page three: a robin, not another person
Adding a small animal expands the cast without crowding it with another figure in similar clothes. A robin works well here. Its warm breast echoes the russet accents, and a damp fence rail repeats the caretaker’s timber.
The link comes from setting and colour, not from giving the bird a scarf or a human expression.
Keep the scale believable. Show enough rail and orchard to explain where the robin is sitting instead of filling the frame with feathers. A closer viewpoint can show the wood grain while distant trees keep the familiar mist.
In the prompt, reference page one again and describe:
- One robin with folded wings, feet gripping a wet timber rail
- The same cool daylight and muted foliage
- The ochre ribbon pushed further back, so it stays a shared detail without competing with the bird’s outline
This page also changes the book’s pace. After two full-length human figures, a small resident draws the reader’s eye to the orchard’s quieter life, widening the world without forcing a plot event.
Building the scenes in CapCut
CapCut’s AI image generator takes both text prompts and reference images, and runs on Seedream 5.0 from ByteDance and Google’s Nano Banana Pro. It returns four results per prompt, which is handy here. You can pick the version whose atmosphere best matches the world bible, then use that one as the reference for every page after it.
A practical order:
- Generate page one from the world description plus the child’s prompt.
- Pick the strongest result, judging the mood and light, not just the girl.
- Use that image as the reference for the caretaker and robin pages.
- Compare the pages side by side before you make more.
When comparing, check three things: the brightness of the sky, the colour of the wet ground, and how light falls on clothing. A sharp golden rim around the caretaker would need a visible reason in this overcast world, and there isn’t one.
Fix problems with short, targeted follow-ups instead of rewriting the whole prompt:
“Keep the caretaker and gate, but soften the sunlight to match the orchard reference and reduce the saturation of the grass.”
Small corrections like this tend to hold everything else steady. For more on why focused edits beat full regenerations, it’s worth reading about an iterative AI image editing workflow. Download each image once its shared details sit comfortably with the rest.
Leave room for the words
Decide where the text will go before every corner fills up with detail. Mist, a distant wall or a stretch of soft grass makes a calm background for type. Dense leaves and high-contrast branches fight small text, however good they look on their own.
The two human scenes deliberately put their figures on opposite sides, so you get different layout options with the same atmosphere. Always test with the actual text, though. An area that looks empty can still have too much contrast behind the letters.
Two print checks people forget:
- The fold. On a two-page spread, keep faces and key clues away from the centre, where the binding swallows them.
- The trim. Printers cut a little off every edge. A ribbon right at the margin may simply disappear, and then it can’t do its job in the story.
Review the whole set at the size readers will actually see.
Before you publish
If the book is going to be sold, two things matter beyond the pictures.
Disclosure. Amazon KDP has required authors to declare AI-generated content, including images, since 2023. Check the current rules on whichever platform you use.
Copyright. The US Copyright Office has said that images generated purely by AI, without enough human creative control, can’t be copyrighted. Your text, your arrangement of the pages, and substantial edits you make yourself can be protected. The raw images may not be. CapCut’s own FAQ notes that generated images typically aren’t copyrighted in the traditional sense and points to its licensing terms, so read those before you print.
The short version
A storybook world that holds together comes down to a few rules: a shared setting, believable light, colours with a job, and one or two recurring details. Write those rules down first. Expand the cast against them, and leave room for the words. Then let the AI explore inside the world you’ve built, not invent a new one on every page.
Related: Images 2.5 AI Image Agent: Is the Prompt Box Becoming Obsolete?
