>>1884
Thank you! :)

With some SD1.5 checkpoints you can achieve at least some degree of consistency for char likeness (at least concerning facial features) by just giving them a random name. Just put something like: "(Irene Clara:1.1)" somewhere within the descriptions of your desired subject.
Sadly as far as I can tell this "feature" that worked very well with older base models has become less reliable with newer models and I don't know if it still works with sdxl/pony at all. My guess is that the closer the model is to the SD1.5 base model the better it works. ...but if it does work with the model you're using it can even help to control your subjects ethnicity and skincolor to some extent (Although not very reliable when using clip skip 2 or higher). A handy tool I'm using is this random name generator: https://www.behindthename.com/random/
Sadly this little SD - gimmick doesn't help that much with the hairstyles so that's been the first thing I had to manipulate and inpaint for almost every pic from that comic thingy.

Paint.net is the 2nd tool I'm using to make all these pics. Its a nice little free alternative to photoshop, etc. Most importantly it can do layers. For example I just sketch some altered details or simply copy-paste some elements (like the orchid in her hair) and then let inpaint have a go at it to nicely blend it in with the rest of the image. I know a1111 also has an inpaint sketch function but for some reason my system almost freezes when I try to use it so I've just stuck to what has worked best for me so far.

Overall I could not have done any of these pics without using some photoshop-esque techniques and inpaint.

For example no matter what I prompted, SD refused to just give me a decent rearview of the girl walking out of her hut with just some jungle background outside. But after a few hundred failed gens I at least had enough to composite something out of them. Took the girl I liked most, removed the shitty background (just with what is available in the "extra" tab in a1111), took two halves of hut interior I liked, a doorway and then grabbed 3 or 4 snippets of jungle for the background. This also made it easy to have the colors of the background washed out to get that sort of blinding effect - also something I couldn't get the AI to do just by prompting in this case. Once the composite was done I just had to use inpainting along the seams of the snippets again.

For the sword stuff it is basicly the same deal. Got some matching pics of the weapon of choice in roughly the correct angles for the subjects poses and camera angles, inserted that and then let inpaint do its magic. In this case with controlnet, either canny (for more finely detailed swords) or softedge (when the weapon is less detailed and/or closer to the viewer and thus bigger).