差異化LoRA訓練，從此以後概念訓練再也難不倒!! 

我用训练的模型测试就有气场，用其他模型测试就没反应，这是泛化性不好导致的吗？

Comfy要確定原因很難~至於沒反應的部分還是要看差距SD1.5的話泛化性比較好，SDXL因為每個模型的的差異比較大泛化可能會比較差而且要看你在哪個模型上練上游還下游

我用的是这个模型训练的，animefull-final-pruned-fp16.safetensors, 因为我找不到你视频里面的提到的模型,搜到的都是二次融合的模型

哈哈哈~我總覺得她又要感冒了怕~

感謝厚愛😄

謝啦~

這個方法其實出來很久了，但是我之前一直在逃避XD

還是要貼錯誤的內容才知道了

@@JackEllie create LoRA network. base dim (rank): 8, alpha: 4.0 neuron dropout: p=None, rank dropout: p=None, module dropout: p=None create LoRA for Text Encoder: create LoRA for Text Encoder: 72 modules. create LoRA for U-Net: 192 modules. enable LoRA for U-Net prepare optimizer, data loader etc. Traceback (most recent call last): File "D:\kohya_ss\train_network.py", line 1009, in trainer.train(args) File "D:\kohya_ss\train_network.py", line 338, in train optimizer_name, optimizer_args, optimizer = train_util.get_optimizer(args, trainable_params) File "D:\kohya_ss\library\train_util.py", line 3389, in get_optimizer value = ast.literal_eval(value) File "C:\Users\huotu\AppData\Local\Programs\Python\Python310\lib\ast.py", line 108, in literal_eval return _convert(node_or_string) File "C:\Users\huotu\AppData\Local\Programs\Python\Python310\lib\ast.py", line 107, in _convert return _convert_signed_num(node) File "C:\Users\huotu\AppData\Local\Programs\Python\Python310\lib\ast.py", line 81, in _convert_signed_num return _convert_num(node) File "C:\Users\huotu\AppData\Local\Programs\Python\Python310\lib\ast.py", line 72, in _convert_num _raise_malformed_node(node) File "C:\Users\huotu\AppData\Local\Programs\Python\Python310\lib\ast.py", line 69, in _raise_malformed_node raise ValueError(msg + f': {node!r}') ValueError: malformed node or string on line 1: Traceback (most recent call last): File "C:\Users\huotu\AppData\Local\Programs\Python\Python310\lib unpy.py", line 196, in _run_module_as_main return _run_code(code, main_globals, None, File "C:\Users\huotu\AppData\Local\Programs\Python\Python310\lib unpy.py", line 86, in _run_code exec(code, run_globals) File "D:\kohya_ss\venv\Scripts\accelerate.exe\__main__.py", line 7, in File "D:\kohya_ss\venv\lib\site-packages\accelerate\commands\accelerate_cli.py", line 47, in main args.func(args) File "D:\kohya_ss\venv\lib\site-packages\accelerate\commands\launch.py", line 986, in launch_command simple_launcher(args) File "D:\kohya_ss\venv\lib\site-packages\accelerate\commands\launch.py", line 628, in simple_launcher raise subprocess.CalledProcessError(returncode=process.returncode, cmd=cmd) subprocess.CalledProcessError: Command '['D:\\kohya_ss\\venv\\Scripts\\python.exe', './train_network.py', '--enable_bucket', '--min_bucket_reso=256', '--max_bucket_reso=2048', '--pretrained_model_name_or_path=D:/stable-diffusion-webui/models/Stable-diffusion/guonifan.fp16.safetensors', '--train_data_dir=C:/Users/huotu/Desktop/fan/yes/img', '--resolution=512,512', '--output_dir=C:/Users/huotu/Desktop/fan/yes', '--logging_dir=C:/Users/huotu/Desktop/fan', '--network_alpha=4', '--save_model_as=safetensors', '--network_module=networks.lora', '--text_encoder_lr=1.0', '--unet_lr=1.0', '--network_dim=8', '--output_name=baoqi_lora_fan', '--lr_scheduler_num_cycles=8', '--no_half_vae', '--learning_rate=1.0', '--lr_scheduler=cosine', '--train_batch_size=3', '--max_train_steps=800', '--save_every_n_epochs=1', '--mixed_precision=fp16', '--save_precision=fp16', '--seed=6666', '--caption_extension=*.txt', '--cache_latents', '--optimizer_type=Prodigy', '--optimizer_args', '--network_train_unet_only', '--optimizer_args', 'safeguard_warmup=True', 'weight_decay=0.1', 'decouple=Ture', 'use_bias_correction=True', 'd_coef=0.5', '--max_data_loader_n_workers=0', '--clip_skip=2', '--bucket_reso_steps=64', '--xformers', '--bucket_no_upscale', '--noise_offset=0.0']' returned non-zero exit status 1.

@@JackEllie 只用 --network_train_unet_only 可以 加了 --optimizer_args "safeguard_warmup=True" "weight_decay=0.1" "decouple=Ture" "use_bias_correction=True" "d_coef=0.5" 就不行

恩...日文叫做コピー機LoRA，因為這個作法最早是青龍做出來的一直叫做複印機法，我不太清楚他的英文