Recently, I adjusted the idea of ​​painting style training, removed the trigger words, used the nai model for training, and used it directly in the nai model. If you use other models, such as anything, you can try to pull the weight to 1.4最近调整了画风训练的思路，去掉了触发词，使用nai模型训练，在nai模型里直接使用即可，如果使用其他模型，比如anything，可以试着把权重拉到1.4————————————————————————————————————————tag: g1a2sweight:0.4-0.8You can also do without tag...,The data set comes from all 15 plot mv, so there are many style scenes, such as the witch's tea party, Inazuma's monster, Fischer's castle, Mona's starry sky,You can give the screenshot of the mv to AI, and let it extract keywords to try to restore the corresponding style and objects, such as teacups, tables and chairs, starry sky, ocean waves, etc.————————————————————————————————你也可以不用触发词 = =，我划拉了所有15个剧情mv，注意只有剧情mv，所以这里面有一堆风格，包括魔女茶会，稻妻妖怪庆典，谢皇堡，莫娜的星空这些，你要是想还原其中的一些特定风格或者物件，可以找到mv，截图并丢给ai让他给你反推关键词，然后你再复原，比如茶会的那些茶具桌椅，星空，海浪啥的

genshin mv style 原神剧情mv风格

gmstyle.safetensors

v2.0

v1.0

通过原神前15个mv训练，虽然大家都只出人但实际包括各种山水数草建筑都有被训练进去权重1-1.4

Files with SHA256: daaf09a37848179ee40b4c6979c99fa8cb11844a33dedd56f8a35e436f440a18