SamuraiTamagoTabemi_サムライたまご食べ美

Overview

This is a Vpred-specific LoRA for Samurai Tamago Tabemi, trained using a derivative model of NoobAI (noobaiVpredItercomp_v1) as the base model.

She is an image character conceived on March 28, 2025, as a promotional character for Samurai Mac's "Aburi Shoyu Fu Tamago Bacon Nikuatsu Beef" (Seared Soy Sauce Style Egg Bacon Thick Beef).

Since it's Vpred-specific, please set the Discrete Sampling Type to v_prediction in "Advanced Model Sampling for reForge" using tools like reForge when generating images.

Datasets

✅Uploaded: Training image dataset.

Prompt

# 🎨 メタ情報 (Artist & Quality)
samurai_tamago_tabemi,
👤 キャラクター基本 (Character Basics)
1girl, samurai, solo,
✨ 身体特徴 (Physical Features)
頭部 (Head)
ahoge, blunt bangs, brown hair, high ponytail, medium hair,
顔 (Face)
black eyes,
体型 (Body)
👘 衣装 (Outfit)
上着 (Topwear)
japanese clothes, short sleeves, striped kimono,
スカート (Skirt)
blue skirt, pleated skirt,
足元 (Footwear & Legwear)
geta, red footwear, tabi, white socks,
装飾品 (Accessories)
badge, bandaid on knee, black gloves, button badge, detached sleeves, elbow gloves, hair ornament, obi, red bow, red headband, fingerless gloves,
🗡️ 武器 (Weapon)
katana, sheath, sword on back,
😊 表情・感情 (Expression & Emotion)
blush, smile,

Training Command

accelerate launch –num_cpu_threads_per_process 4 sdxl_train_network.py –pretrained_model_name_or_path “C:\ProgramTools__GenerativeAI\StableDiffusionModels__SDXL\noobaiVpredItercomp_v1.safetensors” –train_data_dir “H:\Resources\Resources_AI__Dataset\forLeaningImages__Character\Mac_samurai” –output_dir “H:\Resources\Resources_AI__Models__Lola__MyLoRAs\SDXL\Character\samurai_tamago_tabemi” –network_module “networks.lora” –network_dim 16 –network_alpha 8 –network_args “conv_dim=16” “conv_alpha=8” –save_precision “fp16” –xformers –gradient_checkpointing –persistent_data_loader_workers –cache_latents –cache_latents_to_disk –max_data_loader_n_workers 1 –enable_bucket –save_model_as “safetensors” –mixed_precision “fp16” –resolution 1024 –min_bucket_reso 512 –max_bucket_reso 2048 –train_batch_size 3 –max_train_epochs 500 –save_every_n_epochs 10 –optimizer_type “prodigy” –learning_rate 1 –lr_scheduler “cosine” –output_name “LoRA_SDXL_noobaiVpredItercomp_v1_samurai_tamago_tabemi_v3_prodigy_d16a8” –caption_extension “.txt” –max_token_length 225 –keep_tokens_separator=”|||” –shuffle_caption –seed 42 –no_half_vae –logging_dir “E:\Creative\Program\Github__AI\Art\TraningLog\sd_scripts\log” –log_prefix “LoRA_SDXL_noobaiVpredItercomp_v1_samurai_tamago_tabemi_v3_prodigy_d16a8” –fp8_base –optimizer_args “d_coef=0.8” “decouple=True” “weight_decay=0.01” “use_bias_correction=True” “safeguard_warmup=True” “betas=0.9,0.99” –noise_offset 0.05 –v_parameterization –zero_terminal_snr