The children's book referred to here is my The Hat Who Lost Its Head.
As mentioned in The First Ouch, an image AI tends to make variations. When I repetitively asked for a hat, I got different hats. For illustrating this book, the hat images needed to be recognizable as the same hat. The story is about one hat, not many different ones.
The method I adopted is to create a drawing. Then use that drawing as the starting image.
Vectornator installed on a Mac was used to create a suitable drawing of a hat. The drawing was inspired by a public-domain or creative-commons image. The hat drawing was exported to use as an AI starting image. The hat got a tilt and an export was made of that, too. Then one more tilt variation. I now had three visuals of the same hat that could be used as AI starting images.
The hats, as they are, could be used in the book. However, I wanted small variations, the type an AI can provide.
Below are variations of the hat chosen for the cover. The AI art generator at Night Cafe was used.
The last of the above 3 was specified as the start image. The noise weight was reduced to 0%. (The more noise weight, the more "creative" the AI is allowed to be, and I wanted a minimum of creativity.) The text prompt was specified as "hat".
Below are some of the variations generated by the AI. The first variation is likely to be chosen for the cover image. Other variations may be published inside the book.
This book cover is a "first draft," so to speak. But you can see somewhat what I have in mind.
One or another variation of the tilts of the hat is planned to be part of every image in the book.
The images will be generated by the Night Cafe AI art generator instead of the DiffusionBee software on my computer. They both do a good job. But it seems that somehow the Night Cafe one best understands what I need for this project.