split 差分

基于starryXL的更新版本，同样适用于animagineV3.1

要达到图示效果，需同时包含衣物描述和nsfw标签，注意分辨率比例

Updated to the starryXL based version, which also works with animagine V3.1

Note: you should use both the tag of garment and the NSFW

---

触发词split，实现左右差分效果，通常使用960×960正方形图像或宽度略小于高度的图像。

通过服装描述+nsfw可实现表里差分效果。若不添加nsfw，仍会生成两张略有差异的图像，因模型训练时间久远，存在画风影响。

结合角色Lora可实现角色差分，需针对不同Lora调整不同权重，通常应使split权重高于角色Lora权重。补充角色缺失特征时，应通过增加标签描述而非提高Lora权重。

AI可在同一图像中保持角色一致性，差分叠图的相似度通常可达90%以上。

The following is from the machine flip: “Split” trigger word, the effect of left-right difference, generally using 960 square map or width slightly less than height.

Through clothing description + nsfw can achieve table and inside difference. If nsfw is not added, two pictures with subtle differences will also be obtained. Because it was refined a long time ago, there is a style influence.

With the role of Lora, the role difference can be achieved. Different weights need to be adjusted for different Lora. Generally, the weight of split should be greater than that of Lora. To supplement the lost features of the role, tag description should be added instead of increasing Lora weight.

AI can maintain the identity of the character in the same picture. The similarity of differential superimposed images can generally reach more than 90%.