I'll explain the 'SDHK'. This experimental model is fine-tuned on a small number of images outputted by the nijijourney v5 AI against SD1.5.
I recommend you to use 'Hires.fix' and the 'NGH' or 'ngng00' embedding for negative prompts.
Description
FAQ
Comments (6)
感觉没v3惊艳了
V3的颜色和线条非常棒,但是也非常不听话,画手更是灾难.现在看V4感觉有点普通了.我还是喜欢用v3,只能藏手
模型经常不听从prompt指示，多出东西或者少东西。
有时1girl权重加到1.3了还画不出人物来，经常只有背景或者人物只有指甲盖大。
有时背景风格偏3D而人物偏2D，两者并不融合看起来怪怪的，透视也很有问题。
_(:з」∠)_ Here's Bing translation
Models often don't follow prompts, giving more or less things.
Sometimes the 1girl weight is increased to 1.3 and the character cannot be drawn, and often only the background or the character is only the size of a fingernail.
Sometimes the background style is 3D and the characters are 2D, and the two do not blend together, which looks strange, and the perspective is also very problematic.
Sometimes the neck turns green