Tired of AI everywhere? Me too.

Everyone's talking about “slop”: the images and text AI churns out by the ton. Posters, flyers, ads: everything ends up with the same golden lighting, the same smooth faces, and nobody behind it. After a while, you stop looking at any of it. I'm tired of it too.

Except I've worked in tech for twenty years, and I'm the one who put AI at the heart of Toupie, an app that makes children's books. I'm hardly neutral here, and I'd rather say so up front. So I asked myself honestly: what turns an image into slop?

Made for no one

My answer, for now: it's made for no one. Content to fill a slot, something you look at for half a second and forget. And with AI, you can now produce it endlessly, for next to nothing.

Made for someone

A Toupie book is the opposite: it's made for one family. A first day on skis, the new baby coming home, the first night at Grandma's. Stories no publisher would ever print, because their whole audience fits around a kitchen table. And your child recognizes the faces: themselves, their grandparents, their best friend, the dog. My son doesn't look at those pages for half a second. He looks for himself in them.

I'm not trying to replace real books, or the artists who make them. A picture book that took an illustrator months of work will always be more beautiful than what a machine puts out in a few minutes. My kids and I still read them.

But “made for someone” doesn't settle everything when that someone is your child. If you're a parent, you have other questions, and rightly so. I'll take them one at a time, with what I've put in place for each.

Where does my child's photo go?

It's used to draw their character, once. The photo is first checked by a Google tool, then sent to Google Gemini, which does the illustrations, just long enough to create the drawing. It's never saved: Toupie doesn't keep it anywhere. What stays is the drawing. Here's the full path a photo takes.

I use the paid version of Gemini precisely because, on the free version, what you send can be used to train Google's models. On the paid version, Google may keep the image for up to 55 days to detect abuse, never to train its AI. And yes, that means the photo leaves Quebec, to Google, long enough to be checked and turned into a drawing. Quebec's Law 25 regulates that kind of transfer closely, and it's spelled out in the privacy policy. If a company can't tell you clearly where your child's photo goes, that's a bigger red flag than any technical detail.

For your child's friends, I don't even ask for a photo: they're described in words. I don't accept photos of other people's kids: their parents haven't agreed to anything, and I have no way to ask them.

Can the AI produce something it shouldn’t?

This is the fairest worry of all, and I won't brush it off: a model that generates text or images can, in theory, produce anything. That's why nothing reaches you without being checked:

• The photo you upload, before the character is drawn.
• The story idea you write.
• Every edit you make to the text.
• The generated story, before you see it.
• Every illustration, one by one.

The standard is content for ages 3 to 9: no violence, nothing scary, nothing suggestive, no public figures. When a generation doesn't pass, it's retried automatically with stricter instructions, at no extra cost to you. Is it perfect? No, no filter is. At first, mine even blocked “le bateau se met à glisser sur l'eau” (“the boat starts gliding across the water”), about as harmless as a sentence gets. I changed it: now it lets ordinary stories through and blocks what a parent of a 3-year-old would clearly object to.

Who decides the story?

You do. You pick the idea and the characters. The story is written first, and you can read it and edit it before a single image is drawn. If a sentence doesn't sound like you, you change it. Only then does the AI draw. It doesn't talk to your child, it doesn't keep the conversation going, and it doesn't decide anything after the fact. You're the one reading out loud and turning the pages.

That, to me, is what separates it from slop: there's someone behind every book, and it's you.

What the AI still gets wrong

I'd be lying if I told you it's all perfect. The character doesn't always look like your child on the first try, and sometimes you need to regenerate once or twice. The AI handles a concrete activity (“learning to ride a bike”) better than a vague idea (“a fun adventure”). It's recent technology, not magic. Anyone promising you flawless AI is trying to sell you something. I'm trying to sell you something too, mind you. That's exactly why I'm telling you where it falls short.

What it comes down to

When I look at an image, the question I ask isn't “was this made with AI?” anymore. It's “was this made for someone?” For the ads and flyers that all look alike, the answer is no, and I get why people are tired of them. For a book where your child recognizes themselves, with a story you chose and read over, I think the answer is yes. But I'm hardly neutral. Whether it works for you is your call.

Anyway, the ski book I couldn't find exists now. We'll see this winter if it did its job.

Got a question this doesn't answer? Write to me. I'm the one who answers.

— Charles