AnimeXL-xuebiMIX的最新版本，相比之前版本大幅度地改善了图片的光影，让图片层次感和氛围感更强
另外本模型已经融合了LCM-LORA，无需再额外调用，因此使用本模型时推荐以下设置：
1.采样算法选择Euler-a（根据琥珀大佬的意见，实测比LCM采样效果好很多）
2.迭代步数6-10即可产出高质量图片，大大加快出图速度
3.高分修复算法可自行选择尝试，迭代步数8-10为佳，重绘幅度根据算法不同自行尝试
4.CFG Scale设置3-4为佳
当然，这模型也并非完美，目前主要情况是肢体细节仍然不太理想，正在研究改进，希望大家能多多测试返图，谢谢
Description
FAQ
Comments (23)
Looking good so far. Here's a translation in English from what they wrote:
Sure, here is the translation of the Chinese text you provided:
The latest version of AnimeXL-xuebiMIX has significantly improved the image's lighting, making the image's layering and atmosphere stronger.
In addition, this model has already integrated LCM-LORA and does not need to be called separately. Therefore, the following settings are recommended when using this model:
The sampling algorithm should be set to Euler-a. (According to Amber's opinion, the actual test results show that Euler-a sampling is much better than LCM sampling.)
A quality image can be generated with 6-10 iterations, which greatly speeds up the rendering speed.
The high-resolution repair algorithm can be selected and tried on your own. 8-10 iterations are recommended. The redrawing range should be tried on your own depending on the algorithm.
A CFG Scale of 3-4 is recommended.
Of course, this model is not perfect. The main problem is that the details of the limbs are still not ideal. We are currently working on improvements. We hope that everyone can test and provide feedback. Thank you.
希望能有沒融LCM的版本，LCM快了，小細節會丟失無法改善。品質換速度感覺還是很虧。
实际测试加了LCM细节反而增加了，原理不明
为什么我出的图都很灰而且带有很多白点？
vae和采样方法是否选择正确？
@XBX 没有设置VAE，我也出现了同样的问题，这个要配合什么VAE使用吗 佬
@waiceamson sdxl_vae_fp16fix
設置完全沒有問題，也全部跟著大佬的來填，結果一樣是上了一層灰和很多白點點TT
@brook723 VAE是否用正确了？
@XBX 是的。完完全全一摸一樣
@brook723 按理来说不应该啊，你有兴趣的话可以加这个Q群256285020，很多人都在里边，可以互相交流一下
@XBX 佬，有新群号吗，群昨天没了
大佬，能否分享点XL的lora训练经验。根据大佬你的lora参数跟着调依旧一坨屎。。。大佬能不能教教
你用的哪个脚本训练？
@XBX 之前是秋叶的，但老是爆NaN, 最近在尝试用autodl上t4wefan的那个脚本
@andandeai0608 可以试试青龙的那个，我一直用的
@XBX 大佬，有推荐的底模么？还是说就用sdxl1.0_vae0.9那个？
@andandeai0608 琥珀的kohaku或者Animagine XL V3 都可以试试
大佬，元神的这些lora是自创的吗，在这网站好像找不到~~
有些是我自己训练的已经上传了，有些是另外一个人训练的，你可以点开图片找到LORA的出处
The image has a grayish layer and white spots. I have tried changing the VAE, but it didn't fix the issue. What should I do?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.