
LoRA for expanding available pool of emotions of character.
V2 info:
Trigger word - "eexpr"
All preview images were generated on DivineAnimeMix and clip 1.
Recommended lora weight: 0.75-1.
Expression words sometimes must be used with more weight (like 1.1-1.2)
Well working expressions:
embarrassed, evil smile, smile, constricted pupils, half-closed eyes, wavy mouth, shaded face, disgusted
Somewhat working (in order to get to work, must be used with other triggers):
gesugao, shocked, scared
Example prompts you can find in Preview images.
Trained on 294 images of next content:
-
constricted pupils - 71 images
-
gesugao - 25 images
-
wavy mouth - 25 images
-
disgusted - 21 image
-
embarrassed - 23 images
-
evil smile - 21 image
-
expressionless - 21 image
-
scared - 21 image
-
shocked - 22 images
-
smirk - 21 image
-
smug - 23 images
Currently being tested on other models and clipskip levels.
V1 info:
Supported expressions:
bored, angry, embarrassed, expressionless, dead eyes, empty eyes, blush, full face blush, smirk, smug, smile, wide smile, evil smile, grin, constricted pupils, sweating, surprised, shocked, wavy mouth, dot mouth, half-closed eyes and some more.
Better be used in ADetailer with weight of 0.5-0.8 and can be used in general rendering at weight 0.3-0.5.
All prompts you can find in images attached to preview.
All images provided was upscaled via highres fix 2x and passed through ADetailer.
Would love to see reasonable criticism and any recommendations!
描述:
Retrained from zero on 11 concepts, though, not all of them were learned well
训练词语: eexpr
名称: eExpressions2.safetensors
大小 (KB): 73923
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success