推荐底模:Dark Sushi Mix大颗寿司Mix
推荐采样方式:DPM++ 2M Karras
推荐步数:20-30
推荐LoRA参数:<lora:mbti-charaters: 0.8>
开启面部修复,先不要开高分辨率修复,直到你做到期待内的图再发送到后期修复即可。
该模型使用16MBTI人格的图片进行了训练,对发饰、头发、神情(包括嘴巴是张开还是闭合)、衣服、裤子、鞋、动作、手上拿着的东西和周围的氛围都能很好的高度定制。但是该模型对关键词更敏感,所以在做图时需要准确地描述图片后,再用Controlnet,效果会更好。
另外,在训练的时候,为了提高MBTI的代入感,我们对每种人格分析并筛选了最具代表性的特征,你可以选择在对应的人格下输入对应的关键词,即可在还原照片特征的同时,提高你的MBTI特征!
ISFJ:careful, closed mouth, standing, ceremonial gesture
ISTJ:serious, closed mouth, holding a grey book in his right hand, holding a black pen in his right hand, thinking, standing
ESFJ:opened mouth, smile, running, lean forward
ESTJ:smile, closed mouth, holding a blue ruler, standing
ENFJ:closed mouth, smile,holding a sword, standing
ENFP:opened mouth, smile, standing, hello, lifting left hand
INFP:hesitate, closed mouth, dancing, butterflies
INFJ:closed mouth, serious, holding a green magic wand on left hand, standing
ESFP:opened mouth, smile, holding a yellow microphone in left hand, standing, lifting right foot
ESTP:opened mouth, smile, sun glasses, single shoulder bag, standing, put finger forward
ISTP:two hands raise a gun, carry a hammer on the back
ISFP:closed mouth, yellow headband, holding a yellow color palette in right hand
ENTJ:serious, slightly open mouth, holding magic wand in left hand
ENTP:smile, voluble, closed mouth, a plateform, a podium mic
INTP:professional, perform an experiment, closed mouth, holding a chemical medicine in left hand
INTJ:right hand on the back, left hand touches his jaw
Recommended base mold: Dark Sushi Mix Large sushi Mix
Recommended sampling method: DPM++ 2M Karras
Recommended number of steps: 20-30
Recommended LoRA parameters: <lora:mbti-charaters: 0.8>
Open the facial repair, do not open the high resolution repair, until you achieve the expected image before sending it to the later repair can be.
The model was trained using images of 16MBTI personalities, and was highly customizable for hair accessories, hair, expressions (including whether the mouth is open or closed), clothes, pants, shoes, movements, what is being held in the hands, and the surrounding atmosphere. However, the model is more sensitive to keywords, so it is better to use Controlnet when you need to accurately describe the image.
In addition, during the training, in order to improve the sense of MBTI substitution, we analyzed and screened the most representative characteristics of each personality. You can choose to enter the corresponding keywords under the corresponding personality, you can restore the photo features at the same time, improve your MBTI characteristics!