<h3 id="darling~-what-took-you-so-long-to-make-a-lora-of-me">Darling~, what took you so long to make a LORA of me?</h3>I'm proud to finally get that Miia model out once and for all, something I probably should have done for Illustrious back when I started making Monster Musume models. Better late than never, I suppose.For some reason, I never made a lora of Miia for Illustrious, which probably stems from my habit of wanting to "fill in the gap", so to speak. This is the reason why I never made a model for Papi, Rachnee, Centorea and Zombina for Illustrious. Needless to say, I probably will just do whatever I want from now on.Anyway, I decided to go a little lighter on the alts and artstyle tags this time around. That's coming later with merges for the gacha costumes and various other outfits, I just included the core images to teach her base appearance as she appears in the anime and her main outfit. Trust me, I tried training everything at once, it worked poorly, so I settled instead on just doing one or two datasets at a time and then merging them. That should help let me get everything I want out of this lora, even if it'll take a few training sessions to finalize it all. For now, you have the base version, which should be more than enough for the main outfit as well as nudes.Main trigger: mmMiiaAppearance: red hair, amber eyes, slit pupils, red face markings, pointy ears, red ears, golden D hair clips, hair between eyes, long hair, side ahoge, large breasts, lamia, red snake tailMain outfit: tan shirt, front-tie shirt, midriff, cleavage, blue skirt, plaid skirt, pleated skirt, brown beltRecommended weight is 0.7, but it works fine at higher weights as well.

Miia (Monster Musume no iru Nichijou) [Anima]

This second version sacrifices some anime style accuracy in favor of adding more official outfits into the mix. Trainef across two runs of 1500 and 1200 steps respectively, in which the first one is first and foremost trained on a dataset of specifically outfits with 1 to 2 reference images in the full dataset, whereas in the second one I added the core dataset from the first version as well as outfits from the gacha. The result is a little less accuracy when it comes to replicating the main outfit and anime artstyle that the core version of my model provides, but it also means it is more style agnostic and can do her other outfits (mostly) accurately, making it a more viable alternative if you want to experiment with artstyles while working with official outfits. My weight recommendation here is lower as a result (around 0.7-0.8), but beyond that you should be able to work it just fine. I may do some more training in the future on the model to maybe see if it improves any further, but for now it should work out rather well.

Files with SHA256: b6c8af0bfbedcec676cbb65a3f4fc82d14c3b0de794252b78fd04f54cf6f72f0