
Le Terrible | ル・テリブル | 可怖 LoRA
I finally managed to get one right with this character. I wonder if I should try make one for L'Indomptable.
Weight:
Recommanded weight 0.8-1.0 (I only tested it at 1.0)
You won't have to use a lower weight than 1.0 unless you want to use multiple LoRAs
Recommended prompt:
Add one_eye_covered, eyepatch, flower_eyepatch
to generate this character with her eyepatch. I haven't trimmed those tags from the dataset to allow this LoRA to generate this character without that eyepatch.
Skins/outfits:
Default outfit:
LeTerribleDefault, blue_bowtie, breasts_apart, center_opening, clothing_cutout,gauntlets, kepi, vambraces, white_background, white_dress, short_over_long_sleeves, bodystocking_under_clothes, bob_cut, eyepatch, pom_pom_(clothes), short_dress, armored_boots,
Swimsuit skin (Sunlight, Seaspray, and Sanctity):
LeTerribleSwimsuit, blue_bikini, eyepatch, flower_eyepatch, frilled_bikini, long_sleeves,necklace, off_shoulder, one_eye_covered,polka_dot_bikini,ribbon,white_jacket, crown_hair_ornament
Nightgown skin (Silence, Slumber, and Solace):
Cancelled:
-
There is not enough drawing of it (huge risk of overfitting and pose fixing)
-
It tend to fuse with other skins/outfits
Thoses drawing are still in the dataset but that LoRA was not trained with them.
Model used for the example:
V5: YesMix v1.6
That character is owned by Manjuu Games and Yostar.
描述:
40 images
10 repeats
6 epochs
2400 steps (batch size:1)
scheduler: cosine
Text encoder learning rate: 1e-5
Unet learning rate: 0.001
40 images
10 repeats
6 epochs
2400 steps (batch size:1)
scheduler: cosine
Text encoder learning rate: 1e-5
Unet learning rate: 0.001
训练词语: LeTerribleDefault,LeTerribleSwimsuit,one_eye_covered,flower_eyepatch,eyepatch
名称: LeTerribleV5.zip
大小 (KB): 15067
类型: Training Data
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
名称: Le_Terrible_D8_V5_e6.safetensors
大小 (KB): 9338
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success