[Illustrious] Mint Fantome

Disclaimer

• Do not use my LoRA to produce AI image and tagging her fanart hashtag in Twitter/X or if you have an enough contribution in that image then fine.

Description

• WIP LoRA

• Experimental A short viral vtuber with red-dress karaoke (but not available now)

• all images is generated with WAI-IL (832x1216 + Hi-Res Fix 2x)

trigger word

debut costume (remove some component for upper body, cowboy shot or others camera angle

mint fantome, pointy ears, multicolored hair, gradient hair, blue eyes, ahoge, green hair, white hair, short hair, hair ornament, bow, two side up, maid headdress, dress, brwon bow, frills, triangular headpiece, long sleeves, white apron, shoulder cutout, single thighhigh, footwear bow, shoes

any costume

mint fantome, pointy ears, multicolored hair, gradient hair, blue eyes, ahoge, green hair, white hair, short hair, hair ornament, bow, two side up

Limitations

• Default Cosutume

• Alternate Costume is working :)

• No karaoke red-dress costume now.

Training Details

LoRA rank

• LoRA standard

dataset

• 230 images

parameters

• resolution = 1024

• batch size = 3

• network dim,alpha = 16,16

• mix/save precision = bf16/bf16

• optmizer = Lion (weight_decay=0.05)

• UNet LR = 2.5e-5

• TE LR = 7.5e-06

• scheduler = cosine_with_min_lr (min_lr_ratio 0.2)

• l2 loss

steps

• epochs = 5

• total steps = 3834

• repeat = 10

tools

• kohya-ss GUI v24.3.0 FORKED by vjumpkung

• torch 2.5.0 cu124

• Runpod RTX 4090

avg weight

• UNet average weights : 0.010952535254711454

• TE1 average weights : 0.008069108938798308

• TE2 average weights : 0.00657234329264611

*This LoRA is for studying LoRA training with new technique so do not use for damaging the vtuber (also support her too).

*If you see my model reposted on other image generation service please report to me.