改用loha训练缩小了体积提高了清晰度
只需要一个触发词可以触发所有内容
推荐分辨率768×1024
推荐使用Euler a优化器其他优化器效果也差不多
换衣服的时候特征可能会消失把lora强度跳到1.2-1.5就可以解决
大部分二次元模型都兼容
Switching to "loha" for training has reduced the size and improved clarity.
You only need one trigger word to activate all content.
Recommended resolution is 768×1024.
It's recommended to use the Euler A optimizer; other optimizers yield similar results.
When changing clothes, some features may disappear; increasing the "lora" strength to 1.2-1.5 can resolve this.
Most two-dimensional models are compatible.
描述:
训练词语: shenlilinghua girl
名称: xl_dixiya girl-000007.safetensors
大小 (KB): 665272
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
