<h3 id="please-don't-stare-so-much-at-my-eye...-it's-embarrasing.-y27uf48f1">Please don't stare so much at my eye... It's embarrasing.</h3>Today, it is my pleasure to bring you Manako from the anime Monster Musume no iru Nichijou. I was a little bit worried that the model training would have issues properly learning her single eye, but it all turned out well in the end, so there's no issues with the lora.Main tags: mmManako, purple hair, short hair, blunt bangs, purple eyes, monster girl, cyclops, single eye, small breastsMON uniform: jacket, yellow jacket, collar, pauldrons, black pants, elbow pads, knee pads, utility belt, combat boots, black glovesCasual outfit: blue skirt, blue jacket, white shirt, sun hat, hat flower, black pantyhose, white shoes, handbagYukata: yukata, purple yukata, obi, sashSwimsuit: swimsuit, one-piece swimsuit, yellow swimsuit frilled swimsuit, sun hatSuit: suit, black suit, black jacket, black skirt, white shirt, black tie, pantyhose, black shoesWorkout: track jacket, spats, white skirt, pleated skirtAs always, the recommended weight is around 0.8, but I do encourage some degree of experimentation, as always.

Manako (Monster Musume no iru Nichijou) [Anima/Illustrious]

Anima release. Tags are mostly the same as before, just adjusted a bit to better match the outfits. There is data for a couple new outfits, but they did not turn out well, and the later epochs ended up overfitted, so not much I can do about that without significant additions to the dataset (or remove some images that are near identical to each other, I don't need five images where the only variation is the expression). Might do it at a later time, I dunno. Either way, I did a full 10 epoch at the default 1500 steps for 75 images, and was surprised to see it begin overfitting by epoch 6 onwards, so I settled on Epoch 5 since it seemed the most versatile of the bunch. I've tested it and it works just fine at weight 1, but I imagine it only gets more flexible at 0.7.

Files with SHA256: 1a2144e4f995656913baab48b794b8f6799a690ce2e860f6bc240011a084ce3e