[Other][Checkpoint]Uenomigi_DB V1
<p>目前已知的一些问题：</p><ul><li><p>有概率会出现水印，你可以使用类似logo这样的tag来抑制</p></li><li><p>CFG过高可能会导致画面崩坏，个人建议不超过8</p></li><li><p>需要挂载通用VAE</p></li></ul><p>虽然我知道已经有这个LoRA了：<a target="_blank" rel="ugc" href="https://civitai.com/models/6509/uenomigi-90-style-lora">https://civitai.com/models/6509/uenomigi-90-style-lora</a><br />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</p><p>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）<br />我尝试过和几个模型进行分块融合，效果都挺不错。</p><p><br />Translate：<br />Some known problems:</p><p>There is a probability that watermarks will appear. You can use tags like logos to suppress them</p><p>Too high CFG may cause the picture to collapse, and I suggest not to exceed 8</p><p>Universal VAE needs to be mounted</p><p>Although I know this LoRA already exists: <a target="_blank" rel="ugc" href="https://civitai.com/models/6509/uenomigi-90-style-lora">https://civitai.com/models/6509/uenomigi-90-style-lora</a></p><p>But this is my own training ckpt model, which means that you can use it to merge with other ckpt models or merge block weighted</p><p>Personally, I think this effect is better (you can judge the actual situation by yourself)</p><p>I have tried to merge block weighted, and the results are very good.</p>
页:
[1]