Hasumi (Bikini) (Blue Archive) | 羽川莲见（蔚蓝档案）

서문（Preamble）：

俩小馋猫 (´ڡ`)～♡

Two greedy little kitties ᕕ(◠ڼ◠)ᕗ

업데이트 로그（Update Log）：

2025.09.01：

• 백본 모델이 hasumi (blue archive) 및 hasumi (track) (blue archive)를 지원하므로, 이번 릴리즈에서는 수영복 버전의 하스미 LoRA를 제공합니다.

• As checkpoint now supports hasumi (blue archive) and hasumi (track) (blue archive), so this release offers a bikini-focused Hasumi LoRA.

• 수영복 외의 의상을 사용하려면 트리거 단어 hasumi-bikini를 hasumi (blue archive)로 변경하세요.

• If you want to use outfits other than bikini, replace hasumi-bikini with hasumi (blue archive).

• 주요 트리거 단어: hasumi-bikini, black-pendant choker, black-micro bikini, black-low wings, black-thigh strap.

• 보조 트리거 단어: single side bun, side ponytail, eyes visible through hair, red eyes, slit pupils, front-tie bikini top, side-tie bikini bottom.

• Primary trigger word: hasumi-bikini, black-pendant choker, black-micro bikini, black-low wings, black-thigh strap.

• Secondary trigger words: single side bun, side ponytail, eyes visible through hair, red eyes, slit pupils, front-tie bikini top, side-tie bikini bottom.

학습 데이터셋（Training Set）：

이미지 수 = 120

• 2_hasumi-bikini = 120

정규화 이미지 수 = 100

• 1_1girl = 100

에포크 = 27

[Illustrious XL]Hasumi (Bikini) v2[LoRA][REG]：

model_train_type = "sdxl-lora"
pretrained_model_name_or_path = "D:/AI Files/SD-Trainer/sd-models/Illustrious XL 1.0.safetensors"
train_data_dir = "D:/AI Files/SD-Trainer/train/Hasumi (Bikini)/train"
reg_data_dir = "D:/AI Files/SD-Trainer/train/Hasumi (Bikini)/reg"
prior_loss_weight = 1
resolution = "1024,1536"
enable_bucket = true
min_bucket_reso = 512
max_bucket_reso = 2048
bucket_reso_steps = 64
bucket_no_upscale = true
output_name = "[Illustrious XL]Hasumi (Bikini) v2 EP"
output_dir = "D:/AI Files/SD-Trainer/output/LoRA/[Illustrious XL]Hasumi (Bikini) v2[LoRA][REG]"
save_model_as = "safetensors"
save_precision = "bf16"
save_every_n_epochs = 1
save_state = false
max_train_epochs = 30
train_batch_size = 2
gradient_checkpointing = true
gradient_accumulation_steps = 60
network_train_unet_only = true
network_train_text_encoder_only = false
learning_rate = 1
unet_lr = 1
text_encoder_lr = 1
lr_scheduler = "constant"
lr_warmup_steps = 0
optimizer_type = "Prodigy"
network_module = "networks.lora"
network_dim = 8
network_alpha = 4
randomly_choice_prompt = false
positive_prompts = "(masterpiece, extremely aesthetic, best quality, amazing quality), newest, highres, absurdres, cowboy shot, from side, looking at viewer, looking to the side, 1girl, hasumi-bikini, shiny sky, wet, red halo, single side bun, side ponytail, raised inner eyebrows, red eyes, slit pupils, mole under eye, blush, parted lips, black-pendant choker, arms up, armpits, arms behind head, black-micro bikini, front-tie bikini top, side-tie bikini bottom, huge breasts, black-low wings, navel, ass, thighs, black-thigh strap, outdoors, day, blue sky, cloud, simple bird, beach, ocean"
negative_prompts = "(worst aesthetic, worst quality, low quality), low details, bad quality, poorly drawn, (logo, watermark, signature, artist name), (bad hands, mutated hands), extra digits, fewer digits, bad feet, multiple views, 4koma, furry, anthro"
sample_width = 1024
sample_height = 1536
sample_cfg = 5.5
sample_seed = 2333
sample_steps = 28
sample_sampler = "euler"
sample_every_n_epochs = 1
log_with = "tensorboard"
log_prefix = "[Illustrious XL]Hasumi (Bikini) v2 "
logging_dir = "./logs/LoRA"
caption_extension = ".txt"
shuffle_caption = true
keep_tokens = 5
max_token_length = 255
seed = 1919810
mixed_precision = "bf16"
xformers = true
lowram = false
cache_latents = true
cache_latents_to_disk = true
persistent_data_loader_workers = true
optimizer_args = [
"decouple=True",
"weight_decay=0.01",
"use_bias_correction=True",
"d_coef=2.0"
]