Dreambooth 训练 专家模式
你所热爱的 就是你的参数
model_train_type = "sd-dreambooth"
pretrained_model_name_or_path = "./sd-models/model.safetensors"
v2 = false
train_data_dir = "./train/aki"
resolution = "512,512"
enable_bucket = true
min_bucket_reso = 256
max_bucket_reso = 1024
bucket_reso_steps = 64
output_name = "aki"
output_dir = "./output"
save_model_as = "safetensors"
save_precision = "fp16"
save_every_n_epochs = 2
max_train_epochs = 10
train_batch_size = 1
gradient_checkpointing = false
learning_rate = 0.000001
learning_rate_te = 5e-7
lr_scheduler = "cosine_with_restarts"
lr_warmup_steps = 0
lr_scheduler_num_cycles = 1
optimizer_type = "AdamW8bit"
log_with = "tensorboard"
logging_dir = "./logs"
caption_extension = ".txt"
shuffle_caption = true
weighted_captions = false
keep_tokens = 0
max_token_length = 255
seed = 1337
clip_skip = 2
no_token_padding = false
mixed_precision = "fp16"
xformers = true
lowram = false
cache_latents = true
cache_latents_to_disk = true
persistent_data_loader_workers = true