Micro Bikini

description

• all of sample image made it Flux.1 Dev GGUF Q4

• recommended weight : 0.8 - 1

Trigger Word

• [color] micro bikini

Training Details

Dataset

• yellow micro bikini 10 images

• red micro bikini 18 images

• purple micro bikini 9 images

• pink micro bikini 17 images

• orange micro bikini 15 images

• green micro bikini 21 images

• blue micro bikini 16 images

parameters

• resolution = 1024

• batch size = 1

• dim,alpha = 32,32

• mix/save precision = bf16/bf16

• optmizer = AdamW

• UNet LR = 5e-5

• TE LR = 2.5e-5

• scheduler = cosine_with_min_lr (min_lr_ratio 0.2)

• l2 loss

steps

• epochs = 7

• total steps = 2030

• repeats = 4,2,5,2,3,2,3

tools

• kohya-ss GUI v24.3.0 (Forked by me)

• torch 2.5.0 cu124

• Runpod RTX 4090 24 GB + xformers + gradient_checkpointing

weight

• UNet average weights : 0.003128348355459278

• TE1 average weights : 0.007201075088232756

*This LoRA is for studying LoRA training with Flux

*REPOST model is not allow if you found this model please report to me.