一个沙雕动画角色的lora，我训练完跑图的时候才想起来，这200张训练集里居然没有虾仁...

不过现在看泛化倒是挺好的，训练集全部都是古装的服装，用gemini写的泛化提示词，自己试吧，效果还行。

Here is a LoRA for a goofy/meme-style anime character. I didn't realize until I started testing the model that I completely forgot to include any combat/killer poses (or 'shrimp', as we say in slang) in the 200-image training set...

However, the generalization turned out to be surprisingly good. Even though the dataset consisted entirely of ancient Chinese costumes, I used Gemini to generate the variation prompts. Feel free to try it out yourself—the results are actually pretty solid.