This LoRA is intended to produce images of conjoined twins where most of the visual body is shared.Helpful preps to get hits more often might be ...--------------------------------------- Conjoined twins, we can see where their necks meet at the shoulder in the middle, of course the shoulder is missing and it's just one combined chest at that region. ---------------------------------------Or maybe clarify with something simpler...--------------------------------------- There are two of them, conjoined twins, separated at the shoulders upwards. ---------------------------------------All of the training images were synthetic and had to actually be fabricated manually, compositing parts together.Even Illustrious didn't give me joy and the hits were too far away from each other to be a useful time of my life spent.So, I stitched some synthetic people together a this way and a that way, trained a LoRA with some basic words and brought it over to Z-Image. I had horrible body deformities, lines, clean cut areas but then, success! Every once in awhile I'd get a good one and that went into my training data.I varied the nationality, hair color and style, clothing and angles and after a few hours I had 52 images that were ready for preparation. For these I masked out the areas I didn't want the model to learn and sent it over to OneTrainer to produce another Z-Image LoRA.Full body shots are a challenge. I've gotten it to work with some clothing, actually by not being specific about clothing, but most of the training images were head and shoulders to limit the amount of work I had to do, a few were upper body but that tired me quickly and Z-Image seems to know what this is supposed to look like at full body but doesn't get it often. If there's interest in this I'll compose some full body images and see where it goes.

Files with SHA256: 7851e30174c7c5603165b0074a4d7e19e0d8fdf2dc00f9152bf425b2dd840632