凡人修仙传宋玉。基于虚幻5打造的建模，国产建模天花板了

Song Yu from The Immortal Ascension —featuring a character model built on Unreal Engine 5—represents the absolute pinnacle of domestic 3D modeling.

1. 触发词 / Trigger Words

songyu

推荐搭配提示词：

Recommended prompt tags:

(hanfu boots),green dress,white dress,long hair,sash,jewelry,brown black hair,earrings,detached sleeves,sash,necklace,hair ornament,clothing cutout

2. VAE / VAE

使用 SDXL VAE，可在 Hugging Face 搜索下载 sdxlVAE_sdxlVAE.safetensors。

Recommended VAE: search Hugging Face for sdxlVAE_sdxlVAE.safetensors.

3. 数据集说明 / Dataset

训练集为 600+ mixed dataset，混合 face、outfit、head ornament、hair、full-body silhouette 等多种构图。数据集质量决定 90% 的 LoRA 效果。

The training dataset contains 600 mixed images, covering face, outfit, head ornament, hair, and full-body silhouette. Dataset quality determines 90% of the final LoRA quality.

4. 训练参数 / Training Config Toml

model_train_type = "sdxl-lora"
pretrained_model_name_or_path = "path to your base model"
vae = "path to sdxlVAE_sdxlVAE.safetensors"
train_data_dir = "path to dataset"
resolution = "1024,1024"
enable_bucket = false
output_name = "songyu_illustrious"
output_dir = "path to output lora"
save_model_as = "safetensors"
save_precision = "bf16"
save_every_n_epochs = 2
save_state = true
save_last_n_epochs_state = 2
max_train_epochs = 18
train_batch_size = 2
gradient_checkpointing = true
network_train_unet_only = false
network_train_text_encoder_only = false
unet_lr = 0.00006
text_encoder_lr = 0.000006
lr_scheduler = "constant_with_warmup"
lr_warmup_steps = 100
optimizer_type = "AdamW8bit"
network_module = "networks.lora"
network_dim = 96
network_alpha = 96
caption_extension = ".txt"
shuffle_caption = true
keep_tokens = 1
mixed_precision = "bf16"
xformers = true
cache_latents = true
cache_latents_to_disk = false
persistent_data_loader_workers = true
sample_width = 960
sample_height = 1440
sample_cfg = 6
sample_steps = 32
sample_sampler = "DPM++ 2M SDE Karras"
sample_every_n_epochs = 2

5. ADetailer 脸部增强
推荐启用 ADetailer Face Detection，并使用相同 LoRA 进行脸部细节修复。推荐参数：DPM++ 2M SDE Karras、1024×1024、Denoise 0.65，ADetailer Prompt 示例：

It is recommended to enable ADetailer Face Detection and use the same LoRA for face enhancement. Recommended settings: DPM++ 2M SDE Karras, 1024×1024, Denoise 0.65. Example ADetailer Prompt:

songyu