I have been having fun making children’s books with my kids using AI. I have been working on some techniques to make them even more personalized. I have recently been building some of the images for the books using image-to-image. I will describe a bit of the process and some of the options I use.
Below are some examples of the kinds of image options I get when working from photos to make images for the books.
If you want to make a similar stylized book cover or image, my basic outline is below.
example of generating options while using the huggingface free web ui models
Stable Diffusion Denoising Strength is documented as:
Determines how little respect the algorithm should have for image’s content. At 0, nothing will change, and at 1 you’ll get an unrelated image. With values below 1.0, processing will take less steps than the Sampling Steps slider specifies.
As it states, the closer to zero the more like the original image it will be, by the time it is 1, you are just getting images based on the text prompt you provide. The input image will be ignored. I personally find that
0.4 - 0.6 is the best range for my use cases.
the original image with a few examples of images generated while adjusting the denoising strength
I am looking at ways that I can improve the continuity of the style and look of the characters across pages. If I could seed a book with 3-4 photos of my kids possibly in custom for the theme, could I have all pages make not just a little boy who is an astronaut, but make them all look like the same little boy, etc?
If you want to see any of the completed books, the links below are to the amazon pages for the books. As I am still new to this, the later books, I believe to be better than the first few. I have a couple more that are nearly ready to be published.