moki shota

画师moki画风

示例图随便搞的，我还是更喜欢搞NSFW

关于体型的描述只需要”plump“即可，无需”fat“等提示词，剩下的提示词和生成普通的正太的一致

关于不可描述的部位有专门进行优化训练，效果算是我的模型中最好的一个

训练参数

眼睛和脚还是有点问题的，脚的话...那位画师好像不太爱画

model_train_type = "sdxl-lora" pretrained_model_name_or_path = "./sd-models/0208-000147.safetensors" vae = "./sd-models/vae/sdxl_vae.safetensors" train_data_dir = "./train/moki" prior_loss_weight = 1 resolution = "1280,1280" enable_bucket = true min_bucket_reso = 256 max_bucket_reso = 1080 bucket_reso_steps = 32 bucket_no_upscale = true output_name = "moki" output_dir = "/root/autodl-tmp/output" save_model_as = "safetensors" save_precision = "bf16" save_every_n_epochs = 1 save_state = false max_train_epochs = 30 train_batch_size = 12 gradient_checkpointing = true gradient_accumulation_steps = 1 network_train_unet_only = true network_train_text_encoder_only = false learning_rate = 0.0001 unet_lr = 0.0001 text_encoder_lr = 0.00001 lr_scheduler = "cosine_with_restarts" lr_warmup_steps = 0 lr_scheduler_num_cycles = 0 optimizer_type = "AdamW" network_module = "networks.lora" network_dim = 64 network_alpha = 32 log_with = "tensorboard" logging_dir = "./logs" caption_extension = ".txt" shuffle_caption = false keep_tokens = 3 max_token_length = 255 noise_offset = 0.1 color_aug = false flip_aug = false seed = 1337 mixed_precision = "bf16" full_bf16 = true xformers = true lowram = false cache_latents = true cache_latents_to_disk = true

仅供参考