权重:0.6-1
采样器:euler,euler a,调度器karras、sgm uniform,修脸用karras重绘幅度0.5,放大用euler a,sgm uniform,0.25
推荐采用il写实风,3d风,2.5d风大模型
Recommended Parameters for LoRA
CLIP Skip: 2
(Note: Common values are integers from 1 to 12. "2" is a typical setting for balancing detail and prompt adherence.)
LoRA Weight: 0.6 - 1.0
· Start with 0.7-0.8 for general use. Adjust up to 1.0 for stronger character traits or down to 0.6 for more subtle blending.
Samplers:
· Euler a, Karras/ euler a ,sgm uniform/euler,normal work well.
FOR HIRE FIX:euler a,sgm uniform,0.25
· For Face Fixing: Use the "Hires. fix" or "After Detailer" tool with the DPM++ SDE Karras sampler and a Denoising strength of about 0.5.
Recommended Base (Checkpoint) Models:
For best results,pair this LoRA with base models that are strong in these styles:
· ill Realistic Vision (or similar photorealistic models)
· 3D Animation Style models ill
· 2.5D / Semi-Realistic Anime Style ill
Description
v1
FAQ
Comments (10)
checkpoint ?
go to liblib,find windrealmix
diving series is also fine
could u post some for my gallary bro
大佬,时代在召唤,该炼ZIT的LORA了。前面那个不知火舞的,我试了下,效果不太好,加不加LORA差别不大。至于IL的LORA,最好找个C站有的模型炼,不然难以复现你的例图,有些测试也不好做。建议两个方向:1:真人方向用ZIT炼真人LORA(女优,明星,网红之类的),在B站或者其它平台做充电分享,这个是C站和大多数AI平台的真空,它们不允许真人。2:动漫方向用C站已有模型,炼国漫的人物,现在社区太丰富了,日系ACG的人物开发的差不多了。国漫国风的东西相对少。而且需要上传一个你用得顺手的IL模型,尽量实现例图的复现。
这个不知火舞之前试水的作品,感觉3dzit是真不太行,但是很奇怪的是我之前也搞过几个老师脸模全崩了,过两天重新搞下,前两天正tm'准备研究滑块结果md年前全加班
il的话我训练用的私模,那个私模泛用性很强,但是出图效果都是diving系列最好,我这基本上都有标注用的diving
我的例图也基本上都是diving
哥们加个q吧,1059397352
在这感谢大佬了
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.




