v2:

本次v2更新加强训练了倒立人像、被遮挡面部的识别能力，同时增强了结构一致性，用法参见v1。

经测试，采样器推荐：dpmpp_3m_sde+karras

The v2 update has enhanced the recognition ability of inverted portraits and occluded faces, while also improving structural consistency. For usage, please refer to v1.

recommended sampler：dpmpc_3m_ste+karras

V1:

用这个lora可以直出二次元头接三次元身体的图片，省去了自己PS的时间，也避免了看到转换效果一般的真人面部。训练集使用中英双语，皆可触发lora效果。

中文提示词：把动画角色的身体转换成写实风格，保持头部风格不变。保持体型和原图一致。

英文提示词：Transform the body of the anime character into a realistic style while keeping the head style unchanged.Keep the body shape consistent with the original image.

权重建议：0.7-1

With this lora, you can directly create a picture of a anime head connected to a third dimension body, which saves your PS time and avoids seeing a real face with ordinary conversion effects. The training set uses both Chinese and English, which can trigger the Lora effect.

Chinese prompt: 把动画角色的身体转换成写实风格，保持头部风格不变。保持体型和原图一致。

English prompt: Transform the body of the anime character into a realistic style while keeping the head style unchanged.Keep the body shape consistent with the original image.

Weight: 0.7-1

最后，感谢以下AI内容创作者（排名不分先后）给予的训练集帮助

https://www.pixiv.net/users/84309178

https://www.pixiv.net/users/92969522

https://www.pixiv.net/users/18674580

https://www.pixiv.net/users/92077297

https://www.pixiv.net/users/104182689

https://www.pixiv.net/users/83125362

https://www.pixiv.net/users/45359735

以及其他给予了帮助的朋友。

训练使用的所有素材均为AI生成，示例图片仅作为测试模型泛化性，与训练素材无关。