
This is a test of training a LORA that can reliably generate multiple characters. Since Illustrious can already recreate Pyra and Mythra to a decent degree of accuracy this is mostly just a Glimmer LORA that's trying to preserve some amount of the model's base knowledge. It has issues with outfit assignment in group compositions and overall prompt adherence in multi-character scenes. There is also a great deal of drift in the model's base style depending on the prompt in version 008.
This LORA uses the following tokens:
-
glimmer, hairband, hairpiece, brown hair, brown eyes, ponytail, chest jewel
-
kitharode glimmer, halterneck kimono, white footwear, obi, ribbon, wide sleeves, fingerless elbow gloves, pink skirt, black choker, white belt, thigh pouch, thigh belt
-
bathingsuit glimmer, turtleneck bathingsuit, white suspenders, white heels
-
mythra, very long hair, tiara, yellow eyes, earrings, chest jewel
-
melee mythra, thigh strap, short white dress, white elbow gloves, mythra shoes, high-waist pantyhose
-
radiant beach mythra, thigh strap, white swimsuit, white choker, radiant footwear
-
pyra, bob cut, tiara, red eyes, earrings, chest jewel
-
default pyra, backless shirt, short shorts, thigh boots, gauntlets, black gloves
-
pro swimmer pyra, competition swimsuit, pro swimmer footwear
描述:
Added some fan art to the dataset for style regularization purposes.
训练词语:
名称: aegis illustrious 011.safetensors
大小 (KB): 141449
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success