Every few weeks, a new startup claims to have cracked personalised children's books with AI. Your child as the hero! Custom illustrations in any style! Ready in seconds! The demos look magical. Then you actually try to make one.
What AI Can Genuinely Do Well Right Now
Text generation has become remarkably capable. A well-prompted language model can write a coherent 500-word story with proper narrative arc, age-appropriate vocabulary, and even maintain internal consistency about character details. If you tell it your child has a pet rabbit named Biscuit and loves astronomy, it can weave both elements through a bedtime adventure without losing track.
The quality ceiling is genuinely high. I've seen AI-generated stories that would pass muster alongside mid-tier traditionally published picture books. Not award winners, but solid. The kind of book a child asks to hear again.
Where text excels: speed, personalisation depth, and iteration. You can generate twenty story variants in the time it takes a human writer to outline one. This matters when the goal is matching a specific child's interests, reading level, or emotional needs.
The Illustration Problem Nobody Wants to Talk About
Here's where the gap between demos and reality becomes obvious. Image generation models can produce stunning single illustrations. Put "watercolour illustration of a girl riding a dragon over a moonlit forest" into Midjourney and you'll get something beautiful.
But children's books need visual consistency across 10-20 pages. The same character, same style, same proportions, same colour palette. Current models struggle with this badly. Your protagonist might have brown hair on page one, blonde on page five, and mysteriously gain a different nose by page twelve.
Professional illustrators solve this by developing character sheets, maintaining reference files, and applying years of trained visual memory. AI image generators work probabilistically. Each image is a fresh roll of the dice.
Some companies fake consistency by using extremely constrained styles (flat vector graphics, simple shapes) where variations are less noticeable. Others employ human illustrators to fix AI output. The "fully AI-generated" claim often has asterisks.
The Personalisation Spectrum
Not all personalisation is equal. At one end: mail-merge style insertion. "[Child's name] walked into the forest." This is trivial. It's also not particularly magical for kids over age four who quickly notice the mechanical nature of it.
At the other end: genuine story adaptation. A book that's actually different because your child is terrified of the dark versus one who finds it exciting. A narrative that adjusts its complexity based on reading level. Characters whose personalities complement how your specific child sees the world.
The technology for deep personalisation exists. The hard part is designing systems that capture the right information from parents without requiring a 45-minute questionnaire, then applying it with enough subtlety that the story feels crafted rather than algorithmic.
What's Coming vs What's Here
Multimodal models that understand both text and images are improving rapidly. Consistency techniques like character LoRAs (custom-trained visual models) are becoming more accessible. A year from now, the illustration problem will be significantly more solvable than it is today.
But "a year from now" is not "now." Parents researching AI children's books today should expect: excellent personalised text, variable illustration quality, and results that require some curation rather than being perfect on first generation.
The companies being honest about these limitations are the ones building sustainable products. The ones promising magic are often just showing you their single best output from fifty attempts.
We built Fabled because we wanted our own kids to have stories that actually knew them. Not name-insertion gimmicks, but narratives that reflected who they are. The technology still has rough edges. We've chosen to focus on what works brilliantly now — the storytelling — while being transparent about what we're still improving.