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 config rescale system if you're using A1111

I'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 + Feminine Consistency V1.1 (0.31) + SDXL + SD1.5 custom for the CLIP_G fusion

Autism + Feminine Consistency V2 (0.6) + Confetti Anime for the CLIP_L fusion

SD3 MEDIUM is the core base model

This model CAN:

• Generate text still.

  • 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.

• Generate female forms in a fairly consistent fashion

  • 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.

• Generate considerably more complex contextual situations using the female form than anticipated.

  • 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.

• Generate doll-figured females without genitals, nipples, or reproductive organs.

This Model CANNOT:

• 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.

  • This isn't going to happen with this model. That will require actual training and finetuning.

• Pay your taxes

• Provide you psychological validation

• Make sure y- wait what am I typing?

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.