This isn’t exactly a “How It’s Made” post like I said, but I hope it still offers a bit of insight. Since I often mention redoing pages, I wanted to share one that was nearly finished but ultimately didn’t make the cut.
Looking at the page, I think it’s pretty clear why it didn’t work. It's just too random.
I start with the paneling using raw generations and placeholder text. Most of the time I have a plan of coming from A and going to B for a certain page.


This particular page was relatively easy because most of it could be created through prompting. There are no extra characters that interacts with each other, transformation scene or fancy background*. So i really only had to chose the right images and do the details. For the top-right panel, I used a human, recolored her with fox colors.
But let’s focus on the bottom-right panel. It required a bit more editing than the others and, unfortunately, I can’t reuse it for the new version.
The initial generation (which I no longer have) showed her in a kitchen wearing her maid dress. I removed the background and then the dress by recoloring her body in fox tones. Then I sketched a new room and ran about 50 img2img generations until I was satisfied with the result. But I still wasn’t happy with the background. To fix that, I used the curtains from page 13 and the shelf from page 14, composited them together, and blended everything with img2img. After that I included it into the whole image and smoothened out the edged with img2img.
This was just a small breakdown. Obviously, I can't show or comment on every step that goes into the process, but I hope this rough overview gave you some insight.
Tomorrow, I'll upload the whole thing again, because I've been fixing some minor text errors and reworking a few images.