Henkuma's Art Style For images with detailed background, increase lora's strength.

Henkuma Art Style

bucket_no_upscale = truebucket_reso_steps = 64cache_latents = truecaption_dropout_rate = 0.12caption_extension = ".txt"clip_skip = 2dynamo_backend = "no"enable_bucket = trueepoch = 10gradient_accumulation_steps = 1gradient_checkpointing = truehuber_c = 0.1huber_scale = 1huber_schedule = "snr"learning_rate = 0.0001loraplus_lr_ratio = 16loraplus_text_encoder_lr_ratio = 16loraplus_unet_lr_ratio = 16loss_type = "l2"lr_scheduler = "cosine_with_restarts"lr_scheduler_args = []lr_scheduler_num_cycles = 3lr_scheduler_power = 1lr_warmup_steps = 0.05max_bucket_reso = 3072max_data_loader_n_workers = 2max_grad_norm = 1max_timestep = 1000max_token_length = 150max_train_steps = 3714min_bucket_reso = 512min_snr_gamma = 5mixed_precision = "fp16"network_alpha = 16network_args = [ "rank_dropout=0.05", "module_dropout=0.05",]network_dim = 32network_dropout = 0.1network_module = "networks.lora"no_half_vae = truenoise_offset = 0.1noise_offset_type = "Original"optimizer_args = [ "scale_parameter=False", "relative_step=False", "warmup_init=False",]optimizer_type = "Adafactor"output_name = "Henkuma V1"pretrained_model_name_or_path = "illustriousXL20_v20.safetensors"prior_loss_weight = 1resolution = "1024,1024"sample_sampler = "euler_a"save_model_as = "safetensors"save_precision = "fp16"shuffle_caption = trueskip_cache_check = truetext_encoder_lr = [ 5e-5, 5e-5,]train_batch_size = 5unet_lr = 0.0002wandb_run_name = "Henkuma V1"xformers = true

Files with SHA256: db848e2101de39422fe3a39c3a9b6ec37e799c70b162ca8b6ea87c72eca7efc0