Slingshot Swimsuit

description

• First Flux LoRA

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

• recommended weight : 0.75 - 0.8

Trigger Word

• [color] slingshot swimsuit

Training Details

Dataset

• black slingshot swimsuit 17 images

• blue slingshot swimsuit 15 images

• green slingshot swimsuit 18 images

• orange slingshot swimsuit 16 images

• pink slingshot swimsuit 17 images

• purple slingshot swimsuit 10 images

• white slingshot swimsuit 10 images

parameters

• resolution = 1024

• batch size = 1

• dim,alpha = 32,32

• mix/save precision = bf16/bf16

• optmizer = AdamW

• UNet LR = 1e-4

• TE LR = 5e-5

• scheduler = cosine_with_min_lr (min_lr_ratio 0.67)

• l2 loss

steps

• epochs = 10

• total steps = 2560

• repeats = 2,2,2,2,2,3,3

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 : 0.004571231384353156

• TE1 average weights : 0.007479375422311326

*This LoRA is for studying LoRA training with Flux

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