I was encouraged by a few friends to post the stuff I'd been making, mostly as mementos of fun RP sessions and just to see if I could manage it.
Everything I post went through more or less the same process. First thing I do is generate a base image. I've got Stable Diffusion set up on my computer with a Automatic1111 web interface. Depending on what style and subject matter, I have about half a dozen models I switch between.
First thing I do is try to get some kind of base image to use, generally aim for something that has the right characters in the right pose. Part of this is to figure out if I can manage poses for more complex things. Anything with a solo character is easy. Two characters get more difficult, especially if they're interacting with each other.
After getting some kind of base I like, next step is switching over to inpainting, and tweaking and adjusting small sections of the image until I get the characters looking how I want in the right positions. The hardest part is always where things meet up, holding hands, sitting next to each other, that sort of thing. A lot of times I change up the inpainting prompts and areas to get something to work.
Generally I stick to one model per project, but there are a few I've switched around. I generally don't switch models though, because frequently the pieces don't look like they fit. Like having an oil painted character in the middle of a watercolor painting.
The cleanup process varies depending on the piece. Usually I get most of it with smaller bits of inpainting, but the final step is usually going in with GIMP and tweaking, smudging and airbrushing things at a fine level to smooth out the edges and clean up some of the obvious AI artifacting. By the time I move to Gimp I'll frequently head back and forth between GIMP and Stable Diffusion several times until I'm happy with something. And then go back and edit it at least 5 more times when I see "one more thing" I can't believe I missed.
As an example, my black and white fox pic, https://inkbunny.net/s/3548903, I generated only 8 images to find one that I wanted to use. Normally it takes more. However, from inpainting to editing and back, I have a grand total of 651 images in that series tweaking it to get the little pieces right.
That's it in a nutshell. More than happy to talk more about my process if people have questions
