个人测试lora
AI女性人物，上传的图片都没写触发词，也可使用。
触发词：hanyer
推荐权重：0.8
上传例图都带工作流，可按例生成。
生成远景全身图时，面部可能不太像，可尝试用面部修复。
建议使用一些增加细节的lora使人物更真实。
以下为谷歌翻译：
The following is a Google Translate translation:
Personal testing of Lora AI female character.
The uploaded images didn't have trigger words specified, yet they still worked.
Trigger word: hanyer
Recommendation weight: 0.8
Uploaded example images all include workflows and can be generated according to examples.
The prompt words used are from this：https://civarchive.com/models/2231696/sfw5000-and-qwen-and-z-image-or-5000-portrait-and-selfie-wildcards-for-qwen-and-z-image
When generating full-body images in the distance, the face might not look quite right; try facial retouching.
It's recommended to use some Lora tools to enhance details and make the character more realistic.
Description
FAQ
Comments (3)
这是拿了多少菡云芝的数据炼的hhh
人物脸部确实来源于菡云芝，不过是一开始用SDXL炼的一个失败lora，用它生图时人脸修复出来的一个脸型。我就用这一张脸型图片给豆包，生成了几张不同动作角度的素材图。最终训练只用8张图片。现在应该跟菡云芝关系不大了
@skyone2536 挺好的，我还想炼菡云芝角色，但是动画里面菡云芝的表情太多了，炼出来搞得每次脸总觉得不像