Ok, I'll concede to the bad review; this one is not great, probably due in part to the extra/unnecessary tags added by auto-tagging. I'm leaving it published since it can have some really good results when it's cooperative, but it does bork a lot of generations. Anime/Cartoon/Hentai models seem to work best, but it's very picky about realistic models. hopefully manually captioning and making some other adjustments for v2 will make it more consistent. Makes a huge range of androgynous body types. This was trained on 800+ generated images of figures various combinations of gender traits and styles. The general tag "genderless" was applied to most of the set, and then selective WD14 captioning was used, with most gender-specifying terms removed (1boy, 1girl, muscular male, muscular female, etc.). The WD14 captioning was hit or miss at best, and added a lot of unnecessary captions, but this should respond to most standard tags. So far based on testing, weight 0.2-0.5 is best with most models. Anything over 0.5 is likely to deform. SD1.5 version is much more flexible than the SDXL version, and seems to get along with most models.The SDXL version is less cooperative with prompting and not deforming. I've had the best results on it using the SDXL base.

Gender Less SDXL

Currently testing to see if the XL version is viable, it seemed to overtrain very easily

Files with SHA256: a0b05ff1818c07dbd39d2bad7a9e067bf704ae95a99d5e49a8d05b19d5eeb71e