I’ve been exploring with SD for a while to create children’s picture books. I appreciate how it enables a rookie like me who cannot draw properly to illustrate ideas. I won’t call my books art in any sense, but I indeed am willing to show and read them to my own child. I created them with love.
I’ve completed a series of four picture books that are variations of classic nursery songs, and have just started the second series about science. I got a decent understanding of how to do the following things.
- text2image. Use the proper style is important for the final effect. However, the style file also limits how well the text can be understood. For example, I tried “a child pouring juice into a cup” with a cartoon style. SD could not generate a proper pitcher or the child’s gesture. My guess was that there were few samples close to “a child pouring juice into a cup in cartoon style” in the training set, so the model didn’t know what to do. I removed the style and let SD generate a realistic image (by default) of “a child pouring juice into a cup”. Now it worked. The rest was to use image2image and convert the realistic image to cartoon.
- Outpainting. Expand the image beyond its current boundary. Use a high denoise level is crucial.
- Inpainting. It’s very powerful to modify/add/delete objects to the existing image. It’s the key to create convincing and appealing results. It’s also the vital tool to add a personal touch to the machinery.
Next things on my learning list are controlNet and lora training.
By the way, I also learnt how to publish on Amazon KDP. Page setting for publishing was not very fun, and jargons like “bleeding”, “margin”, etc. were not friendly at the first glance. Yet, I did it with Pages on Mac. It turned out that the tutorial on the KDP website was more than enough to help get things done. I’ve saved the settings as templates that will be used by all my picture books.