creator: Gtonero
If using LoRA characters
It is recommended to use BREAK (color kimono:1.2)
Main tag : Color kimono,
Secondary tag : wide sleeves, long sleeves, obi, sash, floral print,
Extra tag : furisode, yukata,
LoRA str : 1
Prompt str : (Color kimono:1.2),
Denoising str : 0.3 to 0.45
Training Pony LoRA Powershell info (v1.0_Pony) :
resolution = 768
Batch size = 2
network_dim = 8
network alpha = 8
mix/save_precision = BF16
optimizer_type = AdamWschedulefree
optimizer_args_AdamW = weight_decay=0.025
GPU RTX 4060Ti 16GB
Torch 2.2+cu121 + Xformers + gradient_checkpointing
Vram used = 10.6GB
Speed = 1.42sec/it
Weight
UNet weight average strength: 0.010724637827052687
Text Encoder (1) weight average strength: 0.008667910846490115
Text Encoder (2) weight average strength: 0.0069243304464663225
Training Pony LoRA Powershell info (v1.0_SD) :
UNet weight strength: 0.01395
Text Encoder weight strength: 0.009704
