<h3 id="preamble-dequhlbh8">前言（Preamble）：</h3>从可畏那里借了白色晚礼服。I took a white evening gown from Formidable😋<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/1076adaf-58bf-4e32-af41-8085059c88df/width=525/1076adaf-58bf-4e32-af41-8085059c88df.jpeg" /><h3 id="update-log-484fz2jkz">更新日志（Update Log）：</h3>2025.07.10：<ul><li>尝试使用正则化技术制作服饰 LoRA，后续会做些小调整（也许？）</li><li>Attempted to create a clothing LoRA using regularization techniques, with potential minor adjustments to follow (maybe?)</li><li>主要触发词：<code>white evening gown</code>，<code>blue cross-laced slit</code>，<code>blue-white cape</code>，<code>black hair spikes。</code></li><li>次要触发词：<code>see-through veil</code>，<code>white veil</code>，<code>pumps</code>，<code>white shoes</code>，<code>empty picture frame</code>。</li><li>Primary trigger word: <code>white evening gown</code>, <code>blue cross-laced slit</code>, <code>blue-white cape</code>, <code>black hair spikes.</code></li><li>Secondary trigger words: <code>see-through veil</code>, <code>white veil</code>, <code>pumps</code>, <code>white shoes</code>, <code>empty picture frame</code>.</li></ul><h3 id="training-set-qfh0sgprx">训练集（Training Set）：</h3>number of images = 27<ul><li>8_default = 27</li></ul>epochs = 15<h3 id="illustrious-xlevening-gown-(anniversary-memories)-v1lorareg.toml-28s8tg2h8">[Illustrious XL]Evening Gown (Anniversary Memories) v1[LoRA][REG].toml：</h3><pre><code>model_train_type = "sdxl-lora"
pretrained_model_name_or_path = "D:/AIGC/ComfyUI-AKI/models/checkpoints/Illustrious XL v1.0.safetensors"
train_data_dir = "D:/AIGC/lora-scripts/train/Evening Gown (Anniversary Memories)/train"
reg_data_dir = "D:/AIGC/lora-scripts/train/Evening Gown (Anniversary Memories)/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]Evening Gown (Anniversary Memories) v1 EP"
output_dir = "D:/AIGC/lora-scripts/output/LoRA/[Illustrious XL]Evening Gown (Anniversary Memories) v1[LoRA][REG]"
save_model_as = "safetensors"
save_precision = "bf16"
save_every_n_epochs = 1
save_state = false
max_train_epochs = 20
train_batch_size = 2
gradient_checkpointing = true
gradient_accumulation_steps = 13
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, sleepy, 1girl, elf, frieren, long hair, white hair, twintails, black hair spikes, pointy ears, crystal earrings, red gemstone, parted bangs, thick eyebrows, half-closed eyes, green eyes, blush, open mouth, parted lips, bare shoulder, collarbone, white evening gown, blue cross-laced slit, hand up, open hand, yawning, squeans, blue-white cape, white background"
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]Evening Gown (Anniversary Memories) v1 "
logging_dir = "./logs/LoRA"
caption_extension = ".txt"
shuffle_caption = true
keep_tokens = 4
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"
]</code></pre>

Formidable's Evening Gown (Azur Lane) | 可畏的晚礼服（碧蓝航线）

2025.07.10：尝试使用正则化技术制作服饰 LoRA，后续会做些小调整（也许？）Attempted to create a clothing LoRA using regularization techniques, with potential minor adjustments to follow (maybe?)

Files with SHA256: 5b01dd1e9a661133e0cd8138ba9cb1a9acebcfe96e48cb57bda62a42967120a8