Hi, I'm Saro, a granddad and CTO, once chief architect of Fidelity NetBenefits and head of engineering at IDEMIA for TSA PreCheck.
Last summer, as I was reading books with my granddaughter, I had a brainwave that she would love to read books with her pictures in it for the main character. By that point, I had already been a heavy user of Claude and Claude Code had come out.
What it does is simple enough. Upload a photo with one or more (up to 4) characters for the comic (or picture book), tell us who's who, choose a story (or make up your own), choose a style, and click to preview. It takes under a minute to show you the cover page and the story written for you. From you can choose to buy a PDF and/or a hardcopy printed book if you want. The full book is only created after you buy. Pricing starts at $4.99 for PDF only or around $20 for printed books before shipping.
I launched it in August 2025 and had several orders. Customers who purchased were happy with their printed books, but I was reluctant to promote it because maintaining consistency across the book was hard. Per-image costs were steep and the cost of redoing books if needed was prohibitive. Vision models couldn't reliably detect hallucination either. But the biggest impediment to progress on this side project is the enormous amount of time it took to do all the testing. Claude Code was good enough at the time for the coding aspects, but testing was still intensely manual. I also had some setbacks with Claude Code making mistakes between branches and causing havoc.
Things got much better starting in early 2026, with the core Claude models for coding, but more recently, Claude Code got much better with building and executing test plans, particularly with the in-built browser able to do the tough job of QA for all the artwork!
It's still a constant struggle to maintain consistency as newer versions of image models (I use a mix of them for different purposes) come up. Producing good stories out of outlines that people give us is another hard item.
I'd love to get feedback from this tough audience on what they think of the image consistency and the stories, and if they can give any advice on how to improve either.
I want to also address the elephant in the room. Some people won't recommend joonoo.art, because books made with AI hurt the people who write and draw for a living. AI art raises fair questions. We label every book as AI-made and never imitate living artists, and I've written honestly about what we owe artists and authors: joonoo.art/blog/what-we-owe-artists-and-authors. I'd love to hear from you about this too!
cyberviso•10m ago
Last summer, as I was reading books with my granddaughter, I had a brainwave that she would love to read books with her pictures in it for the main character. By that point, I had already been a heavy user of Claude and Claude Code had come out.
What it does is simple enough. Upload a photo with one or more (up to 4) characters for the comic (or picture book), tell us who's who, choose a story (or make up your own), choose a style, and click to preview. It takes under a minute to show you the cover page and the story written for you. From you can choose to buy a PDF and/or a hardcopy printed book if you want. The full book is only created after you buy. Pricing starts at $4.99 for PDF only or around $20 for printed books before shipping.
I launched it in August 2025 and had several orders. Customers who purchased were happy with their printed books, but I was reluctant to promote it because maintaining consistency across the book was hard. Per-image costs were steep and the cost of redoing books if needed was prohibitive. Vision models couldn't reliably detect hallucination either. But the biggest impediment to progress on this side project is the enormous amount of time it took to do all the testing. Claude Code was good enough at the time for the coding aspects, but testing was still intensely manual. I also had some setbacks with Claude Code making mistakes between branches and causing havoc.
Things got much better starting in early 2026, with the core Claude models for coding, but more recently, Claude Code got much better with building and executing test plans, particularly with the in-built browser able to do the tough job of QA for all the artwork!
It's still a constant struggle to maintain consistency as newer versions of image models (I use a mix of them for different purposes) come up. Producing good stories out of outlines that people give us is another hard item.
I'd love to get feedback from this tough audience on what they think of the image consistency and the stories, and if they can give any advice on how to improve either.
I want to also address the elephant in the room. Some people won't recommend joonoo.art, because books made with AI hurt the people who write and draw for a living. AI art raises fair questions. We label every book as AI-made and never imitate living artists, and I've written honestly about what we owe artists and authors: joonoo.art/blog/what-we-owe-artists-and-authors. I'd love to hear from you about this too!