雾都迅羽的触发词为：

Fu Hua(HoF), white shirt, brown coat, purple ascot, belt, black shorts, grey socks, brown boots,

可选：wrist watch, monocle, brown gloves, ponytail, long hair,

在使用”monocle,”时，推荐把”wear monocle,”也同步加入，并将”glasses,”加入negative prompt中（不然有可能出现把单片镜拿在手里，或是变成普通的眼镜的情况）

可以加入 “face,“来获得更贴近角色的脸部。

推荐权重：0.8-0.9

该模型在权重0.8时仍有较好的换衣或nsfw功能

此模型基于“animefull-final-pruned.ckpt”，不同大模型适配程度见图10

如有任何建议，欢迎留言。能有返图就再好不过啦！

For the " Hawk of the Fog" prompt:

Fu Hua(HoF), white shirt, brown coat, purple ascot, belt, black shorts, grey socks, brown boots,

Optional：wrist watch, monocle, brown gloves, ponytail, long hair,

When using the trigger word "monocle," it's recommended to include "wear monocle," in the prompt and include "glasses," in the negative prompt(otherwise, there might be instances of holding the monocle or it turning into regular glasses).

Adding "face," can create a closer match to the character's facial features.

Suggested weight: 0.8-0.9

The model performs well in clothing changes or NSFW functions at a weight of 0.8.

This LoRA is based on the "animefull-final-pruned.ckpt",and the flexibitiy of different checkpoint can be seen in Figure 10.

Any suggestions are welcome. Having image feedback would be greatly appreciated!

EN Trans by ChatGPT

写在最后的碎碎念，这次训练的数据集其实在小半个月前我就整好了，但秉持着每次炼丹都得整点新活，我这次花了好多好多时间在调整参数上。在把几个大佬的视频反复品鉴和显卡的悲鸣后，金字塔噪声总算是成功运用在了这次的训练当中。功夫不负有心人，这次的结果是相当可人的，不单泛用性在0.8就时有很好的效果，画面的明暗也得到了一定的改善（可喜可贺）！

但缺点就是用了金字塔噪声后模型超难拟合！无奈只能大幅调高repeat和学习率了，每炉丹的时间成本又上升了（悲）~