I legitimately advise against using this one. It's not a very good outcome in comparison to the more recent experiments. I won't pull the model as it's a basis for education.CLIP_G and CLIP_L are merged into it already. USE THESE CLIPS OR NOTHING HAPPENS.This is a direct merge of SD3, SDXL, and SD1.5's clips using ComfyUI and Dare merger system.I STRONGLY advise using the <a target="_blank" rel="ugc" href="https://github.com/Seshelle/CFG_Rescale_webui">config rescale</a> system if you're using A1111I'll release my workflow tomorrow when I get up. I'm too tired currently and it needs a facelift before I hand it out. CLIP_G and CLIP_L are merged into it. You must load CLIP_T5/xx if you wish to use it as well, and it adds a lot of refinement if you do.WHY did I do this? SD3 is ass by default. You know it, I know it. I know for a fact that it has some really solid potential, so I'm giving it at least SOME consistency for the female form, while damaging some behaviors. It's a decent experiment that seems fun to play with, generates images rapidly, works a bit with SDXL loras, and even adds a lot of interesting new context to the system that I didn't think would be possible without direct finetuning or image training.The model itself is merged with a merged SDXL merged with SD1.5:Autism + <a target="_blank" rel="ugc" href="https://civarchive.com/models/477291/anime-consistency-v11-loha-pdxl">Feminine Consistency V1.1 (0.31)</a> + SDXL + SD1.5 custom for the CLIP_G fusionAutism + <a target="_blank" rel="ugc" href="https://civarchive.com/models/477291/anime-consistency-v11-loha-pdxl">Feminine Consistency V2 (0.6) </a>+ Confetti Anime for the CLIP_L fusionSD3 MEDIUM is the core base modelThis model CAN:<ul><li>Generate text still.<ul><li>I made sure this merge in particular can handle text, so I tried not to destroy too much of the information and I left enough intact to allow this. It's often hit or miss, but you can definitely make it happen.</li></ul></li><li>Generate female forms in a fairly consistent fashion<ul><li>It could do this before, but now it has a bit more consistent logic behind it, which damages some of it's training and behaviors.</li></ul></li><li>Generate considerably more complex contextual situations using the female form than anticipated.<ul><li>It can do a lot more than I expected, though I didn't really have ANY expectations getting into it. I just started playing with it until something happened.</li></ul></li><li>Generate doll-figured females without genitals, nipples, or reproductive organs.</li></ul>This Model CANNOT:<ul><li>Generate porn, sex, or direct sexual situations. The entire UNET is based on SD3, so if there's any porn in there, it's based entirely on what is IN sd3.<ul><li>This isn't going to happen with this model. That will require actual training and finetuning.</li></ul></li><li>Pay your taxes</li><li>Provide you psychological validation</li><li>Make sure y- wait what am I typing?</li></ul>It does whatever SD3 does with a bunch of extra clip information. Sometimes it's good, sometimes it's terrible. Feminine consistency isn't the most solid model for this sort of thing, but it sure leaves an imprint.

Files with SHA256: 246d27697b37c722c4d531ebb083cd29bdb112cff674dcf087d273316d0f6029