斗破苍穹动漫，萧薰儿中洲服饰Battle Through the Heavens Anime, Xiao Xun'er's Zhongzhou Outfit1.触发词/Trigger Words:<pre><code>&lt;lora:xiaoxuner_zhongzhou_outfit_pony:0.85&gt;,xiaoxuner_zhongzhou_outfit,black hair,1girl, green blue dress,detached collar,clothing cutout,bare shoulders,gold trim,sleeves,hair ornament,earings</code></pre>2. 训练时选用的 <a target="_blank" rel="ugc" href="https://civarchive.com/models/1704957/diva-3d-danira?modelVersionId=2201745&amp;sync-account=green">Diva 3D Danira </a>base mode 无VAE，可搜索Hugging Face下载sdxlVAE_sdxlVAE.safetensorsThe model was trained using the<a target="_blank" rel="ugc" href="https://civarchive.com/models/1704957/diva-3d-danira?modelVersionId=2201745&amp;sync-account=green"> Diva 3D Danira</a> base model (no VAE). You can search on Hugging Face to download sdxlVAE_sdxlVAE.safetensors. 3. 权重0.75-0.85服饰还原度较高，配合adetailer lora<a target="_blank" rel="ugc" href="https://civarchive.com/models/2657810?notOwner=true"> xiaoxuner_zhongzhou_face</a>Using a weight between 0.75 and 0.85 gives higher clothing accuracy. For best results, use together with the adetailer LoRA: xiaoxuner_zhongzhou_face. 4. 训练集参考Optional FilesPlease refer to the Optional Files for the training dataset reference. 5. 完整训练参数，感兴趣的朋友尽情调优Here are the complete training parameters, feel free to fine-tune.<pre><code>model_train_type = "sdxl-lora"
pretrained_model_name_or_path = "F:/SD/sd-webui-linux/models/Stable-diffusion/diva3DDanira_v10NoVAE.safetensors"
resume = ""
vae = "F:/SD/sd-webui-linux/models/VAE/sdxlVAE_sdxlVAE.safetensors"
train_data_dir = "F:/SD/training_dataset/btth/xiaoxuner_zhongzhou_outfit"
prior_loss_weight = 1
resolution = "896,1344"
enable_bucket = false
min_bucket_reso = 512
max_bucket_reso = 1536
bucket_reso_steps = 64
bucket_no_upscale = true
output_name = "xiaoxuner_zhongzhou_outfit_pony"
output_dir = "F:/SD/trained_loras"
save_model_as = "safetensors"
save_precision = "bf16"
save_every_n_epochs = 2
save_state = true
save_last_n_epochs_state = 2
max_train_epochs = 24
train_batch_size = 2
gradient_checkpointing = true
gradient_accumulation_steps = 1
network_train_unet_only = false
network_train_text_encoder_only = false
learning_rate = 0
unet_lr = 0.00008
text_encoder_lr = 0.000005
lr_scheduler = "constant_with_warmup"
lr_warmup_steps = 50
optimizer_type = "AdamW8bit"
network_module = "networks.lora"
network_dim = 64
network_alpha = 64
randomly_choice_prompt = false
positive_prompts = "xiaoxuner_zhongzhou_outfit, 1girl, long hair, solo, dress, full body, jewelry, hair ornament, earrings, bare legs, black hair, white dress, railing, detached sleeves, blue dress, brown hair, bare shoulders"
negative_prompts = "lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry"
sample_width = 896
sample_height = 1344
sample_cfg = 6
sample_seed = 2333
sample_steps = 30
sample_sampler = "euler_a"
sample_every_n_epochs = 1
log_with = "tensorboard"
logging_dir = "./logs"
caption_extension = ".txt"
shuffle_caption = true
keep_tokens = 1
keep_tokens_separator = ""
max_token_length = 255
color_aug = false
flip_aug = false
random_crop = false
seed = 1337
mixed_precision = "bf16"
no_half_vae = true
xformers = true
lowram = false
cache_latents = true
cache_latents_to_disk = false
persistent_data_loader_workers = true
vae_batch_size = 1
optimizer_args = [ "weight_decay=0.01" ]</code></pre>

Files with SHA256: 24442f54c48f10f990a8605f893c3f9a8bb2f1f618e04a0cff1b88523e744470