gtsFoot
Does pretty much what it says on the tin. Produces scenes focused around the feet of giant women. Reliably supports a variety of characters, settings, and clothing types. I trained it exclusively on human women, but if you want to try something else, I'm pretty sure Pony can figure it out. Limited control of the placement of other people, but it always seems to give something in the vicinity of what I'm looking for with a success rate of maybe 1/4-1/3 images requiring only minor touchups and maybe 1/8 flawless.As might be expected from something focused around feet, SD will sometimes give you extra or missing toes, inverted feet, merged limbs/feet, or turn feet into horrifying blobs of flesh, but all the included images were straight out of txt2img with no hires-fix or post-processing (sometimes adetailer on the faces, since they tend to be small in the render because of the poses involved). I experimented with some LoRAs and embeddings intending to produce more anatomically correct feet, but it's hard to tell if they helped in any way. Has a weird tendency of rendering the giantess sitting in mid air. Not sure if that's related to this LoRA or a general quirk.Some key pointers: I find 1.0 weight tends to do fine, but I've yet to test in detail. gtsFoot is the intended trigger word. If you don't include "giantess", it tends to give you more of a solo foot pinup.Some things I tried to teach it with mixed success:inSock - this was a bit of a reach. was hoping to capture tiny people trapped inside the giantess' socks/pantyhose/stockings. sometimes it'll give the impression of something in there, though it's not easily resolvable as a person. Might be a good starting point for inpainting.footSandwich - people sandwiched between the woman's feet. Low success rate. Sometimes it gets the idea, but it also sometimes interprets this as putting people between toes and sometimes gets confused about which toe goes with what foot or where one foot ends or another begins.underFoot - people directly underfoot. It'll do that, but I suspect that's a matter of chance anyway.betweenToes - focus on people between and around a giant woman's toes. it generally gets the idea, but it's hard to get the people to not merge with their surroundingsinShoe, xray - An attempt to show a cutaway of the tiny person inside a woman's shoe. Unfortunately it seems to prefer putting a phallus inside certain other places. I suspect that says more about booru tags and PonyXL's training set than anything else. I might retrain with just 'inShoe', since xray is apparently overloaded.Hopefully with this as a baseline I can build a larger dataset and train something more refined.
