一个基于 Niji 的某个提示词生成的风格训练的 Lora,触发提示词“runstyle”,加入之后图像的高光到暗部的过渡会很柔和,同时对于对比色光线的响应会非常好,比如“rose Red and blue, light bronze and indigo,”。
与真人模型混用的话建议权重 0.5,与动漫模型混用建议权重 0.8。
Using Niji-based style training with a trigger word "runstyle", the transition from highlight to dark part of the image will be soft, and the response to contrasting color light will be great, such as "rose Red and blue, light bronze and indigo". When mixed with real-person models, the weight is suggested to be 0.5, and when mixed with animation models, the weight is suggested to be 0.8.