/upscale
masterpiece, best quality, 1girl, (slimthick, massive thighs, hyper hips, gigantic ass:1.4),
masterpiece, best quality, highres, highres, absurdres, haru (re ilust), tokoyami towa, 1girl, alternate costume, arm support, arm up, armpits, ass, bare arms, bare shoulders, bikini, black bikini, black choker, breasts, choker, cleavage, criss-cross halter, demon tail, ear piercing, earrings, fishnet thighhighs, fishnets, frown, green eyes, halterneck, jewelry, long hair, looking at viewer, medium breasts, multi-strapped bikini, multicolored hair, nail polish, navel, o-ring, o-ring choker, open mouth, piercing, purple hair, sitting, solo, stomach, streaked hair, string bikini, swimsuit, tail, taut bikini, thigh strap, thighhighs, thighs, twintails, very long hair, virtual youtuber, wet, yokozuwari
(slimthick, massive thighs, hyper hips, gigantic ass:1.4),
<lora:hyper_bottom_heavy_v3_weighted-lr2e-4-4b-4ga-16dim-flip-te1e-4-tkn225-a8-t_drp0.15-drp0.3-cosine_with_restarts-cap_weight2-n-locon-[bottomheavy]-000140-:1.2>
Similar to my last released hyperbreasts model, this one should be less damaging to the overall style while also making it easier to achieve large sizes.
Also I improved the accuracy of the tags, and doubled the dataset size to ~400 images.
Increasing LoRA strength is actually useful for achieving larger sizes in combination with the right size tags.
Training Details:
~400 images
160 epocs
learning rate 2e-4
text encoder LR 1e-4
base model Av3
clip skip 2
random flip
tag drop chance 0.15
network dropout 0.25
bucketing at 768
dim 32
alpha 16
225 tokens
cosine with restarts
training with tags, tags attached next to model download
use the new weighted captions + dropout in Kohya that way more important tags were trained at a higher weight (weight of:2).
Kohya LoRA LoCon, does not require any additional extensions to use