On my other LoRA's I've stated that I like them to be able to handle more than 1 character. That got me to wondering what else was possible. I often use <a target="_blank" rel="ugc" href="https://civarchive.com/models/8730/hipoly-3d-model-lora">HiPoly</a> in my renders, sometimes one of the MANY different breast size / body proportions LoRA's, then maybe a character LoRA (similar to the Nobody series from <a target="_blank" rel="ugc" href="https://civarchive.com/user/Zovya/models">Zovya</a>) also. Then comes the hassle of finding the right weights of each one to generate an image that's not plagued with fractals/artifacts and still produces the desired result.My answer to this problem was fairly simple really, it just took some time to accomplish. Generate a LOT of images the way I want them, and make a LoRA out of it. 380 renders of 19 different girls and 133 renders of 7 different breast sizes later I've com up with MIO. (many-in-one) About 99% of the training images were created using <a target="_blank" rel="ugc" href="https://civarchive.com/models/22922/lyriel">Lyriel</a> so it will work quite well with it. It has been "baked" slightly more than I prefer but that was because of the way I accomplished the end result. The characters and breasts were trained separately creating a LoRA for each that was a little heavier than I like (about 40% more) then merged 50-50 in Kohya_ss.As I said, there's 19 girls in all and each is slightly different even if sometimes their faces tend to be similar. Most have a clear nationality that comes through in renders even without prompting such as Cheyenne (Native American), Lupita (Latina), Priya (Indian), what I mean by this is that quite often Cheyenne has deerskin clothes, if you put a dress on Lupita it likes to look like a quinceanera gown, and Priya leans towards a sari.On the breast sizing side of things, the basic prompts with an underscore are the correct ones, but to add more precision and get the results as I intended them you should use them in conjunction with helpers. IE when I prompt (flat chest) I'm usually meaning just that, but most checkpoints and breast sizing LoRAs will actually give you what I consider small breasts with that prompt. To fix that just combine (flat_chest) with (no bewbs). Small, medium, and large are pretty straight forward. Huge, giga, and crazy used alone often look almost similar but the intention is for huge to be just that, huge naturals. Giga is more of a huge fake breast, and crazy is well, crazy. The fix is to prompt as (big giga_breasts) and (gigantic crazy_breasts) just be prepared to negative prompt big/wide/thick hips, unless that's what you want.Please like and review. post your creations with it. I love seeing what people make more that seeing the downloads go up.The only thing I ask is to please prompt responsibly. This Lora is geared towards anime/semi-realism and because of what it combines together it REALLY likes to create NSFW images but typically SFW in the positive and NSFW in the negative helps to greatly reduce this to near none.Fun side note Tuya was not actually trained as a furry but because of the hairstyle she had in her training images the end result is a girl that often has dog/wolf ears. if you increase the LoRA weight over 1 and add furry to a prompt with her name you can have some fun with it.

MIO 

While making <a target="_blank" rel="ugc" href="https://civitai.com/models/88608">Millennium Girls</a> I learned a few new things and wanted to revisit MIO. This version I have removed the extra girls added with BAB, I've removed the breast and butt training. This version has training on only the 19 girls originally included with MIO. Later I will make a standalone version of BAB (butts and boobs) or perhaps I will separate them.The reason for this change is that as MIO evolved in the previous versions it became harder and harder to render each girl the way they were intended to look. There had to be more tags included to get the right hair, skin tone, body, etc. This version now has only two purposes, to be a girlpack, and a render enhancer. (For some reason Jun still needs the white hair prompt and Nadia still needs short blonde hair) but all others work much better.The last image is an example of how well it will work to add detail with fewer prompts. The Kate Beckinsale image uses none of the triggers for the LoRA and has no positive quality prompts.

Files with SHA256: ebf58e2f430d0a8a32e61c541c51fb936c46d9a5a536d03a1b260411cfba584b