Today I wanted to figure out the visual style for my first children’s book. I like the style of the books from ‘Muis’, but I don’t want it crossing any legal line.
The business plan pointed me to ComfyUI for day 2. I found out that it is an open source tool for image generation. More importantly it is able to keep character consistency throughout multiple images. Just what I needed for my custom storybook.
However first I wanted to see if I can derive a style that is not the same but still visually appealing for the grandparents that are going to buy these books.
So I fired up Gemini and tried some specific Nano Banana 2 examples for a storybook page. When I prompted for a derivation of the ‘Muis’ style, the result was so good I questioned whether I was looking at an actual page from the book.
I tried again prompting another page in the same book, but turn it into a dark and twisted version instead. To test how far I could push the style. And Gemini complied and the result was stunning. At least now I was sure this wasn’t an actual page from the book. The style held up even in a completely different tone.
Then I tried adding a photo of a random stockphoto of a toddler to see how it would be implemented in the story. Now I was actually shocked to see how good the result was. The photo of a real child worked inside the storybook style. It looked like a real personalized book page.
The next step will be turning this into a pipeline via ComfyUI. Something I will continue with in a future experiment.
Key Insight:
The current AI image generation is good enough to produce near-publishable children’s book pages from a style reference and a photo.