目前已知的一些问题：<ul><li>有概率会出现水印，你可以使用类似logo这样的tag来抑制</li><li>CFG过高可能会导致画面崩坏，个人建议不超过8</li><li>需要挂载通用VAE</li></ul>虽然我知道已经有这个LoRA了：<a target="_blank" rel="ugc" href="https://civarchive.com/models/6509/uenomigi-90-style-lora">https://civarchive.com/models/6509/uenomigi-90-style-lora</a> 但是这个是我自己训练的ckpt模型，也就意味着你可以用它来和别的ckpt模型进行融合或者分块融合个人感觉我这个效果更好（具体实际情况如何你可以自行判断） 我尝试过和几个模型进行分块融合，效果都挺不错。 Translate： Some known problems:There is a probability that watermarks will appear. You can use tags like logos to suppress themToo high CFG may cause the picture to collapse, and I suggest not to exceed 8Universal VAE needs to be mountedAlthough I know this LoRA already exists: <a target="_blank" rel="ugc" href="https://civarchive.com/models/6509/uenomigi-90-style-lora">https://civarchive.com/models/6509/uenomigi-90-style-lora</a>But this is my own training ckpt model, which means that you can use it to merge with other ckpt models or merge block weightedPersonally, I think this effect is better (you can judge the actual situation by yourself)I have tried to merge block weighted, and the results are very good.

Files with SHA256: 29f5834c412ca6feb37557907ca359fce01d33d70afedb9763a4ec9c137f4b7a