这次的主要触发词有三个“danjin”，“braid buns”，“earrings”。基本上这三个用来控制丹瑾的主要特征。其他的部分都可以自由发挥。
由于训练集中很多都是MMD模型出图，不可避免的学到了MMD人物材质的质感。所以经过测试，在提示词的最后加一个“3D to 2D style rendering”，效果会好很多。
推荐的参数范围在0.8-0.85。
测试的图片都是个人的偏好，偏向日常风格。但实际训练集中又很多穿战甲类型的。所以尝试一些加入奇幻元素的穿搭应该也是OK的。
感兴趣的小伙伴可以尝试一下。
There are three main trigger words for this version: “danjin,” “braid buns,” and “earrings.” These three are primarily used to control the key features of Danjin.
Since many of the images in the training set were generated using MMD models, it inevitably learned the texture and material characteristics of MMD characters. Therefore, after testing, I found that adding "3D to 2D style rendering" at the end of the prompt significantly improves the results.
I recommend setting the weight range between 0.8 and 0.85.
The images I tested are based on my personal preferences, leaning toward a daily casual style. However, the training set also includes a lot of armored outfits, so experimenting with fantasy-themed clothing should also work well.
Have fun with it.