My English is not very good, there may be some translation problems, please forgive me.
I have created another very small model, if anyone likes it I will continue to optimize it.
我又来上非常小众的模型了,,, 如果有人喜欢我会继续优化。
首先感谢@遇见清风 提供素材,与我共创。
-------------------------------------------------------------------------------------------------
[Model introduction and parameter recommendations]
trigger : skeleton
The recommended weight of LORA is around 0.6
It is suitable for simple prompt words, model layout, realistic model or two-dimensional animation. The twomodels used in the example pictures are these two: Dream_and_Daily 梦与日常, Dream_and_Fantasy 梦与幻想. Can also be paired with other similar models.
The sample diagram has complete parameters! Everyone is welcome to try it!Welcome to return pictures! Thanks.
My common parameters
Sampler: DPM++3M SDE Karras
Number of steps: 30
Hire Upgrader: R-ESRGAN 4x+
Denoise intensity: 0.5,
Clip skipped: 2,
Hire upscale: 2,
Hire steps: 20,【模型介绍及参数建议】
可以产出骨架身体~ (显而易见)
触发词是:skeleton 骨架
LORA 权重 建议是0.6左右
适用于简单的提示词,大模型搭配写实模型或二次元动漫都可以。
例图用到的大模型是这两个:Dream_and_Daily 梦与日常、Dream_and_Fantasy 梦与幻想。
也可以搭配其他大模型。
---------------------------------------------------------------------------------------------
例图上都有完整参数!欢迎大家尝试!
欢迎返图!感谢。
还是在这里放上我例图用了最多的参数
尺寸:512x768 (为了速度我一般只开这么大,然后开高清修复)
Sampler:DPM++3M SDE Karras
Steps:30
Hires upscaler: R-ESRGAN 4x+
Denoising strength: 0.5,
Clip skip: 2,
Hires upscale: 2,
Hires steps: 20,
欢迎感兴趣的朋友一起学习交流,VX:xat_0313 、XHS:嚣思_Rylee
Description
FAQ
Comments (6)
大佬你好,我想请教一下,这种lora是如何训练出来的,在训练集部分,是用只有头部的图片和只有身体骷髅的图片来训练的吗,谢谢
全身、半身都有
@XS_Rylee 你的意思是说,在训练集中,每张放入的图片都包括人头和骷髅身体吗,还是说会把人头和骷髅身体放在不同的图片里面,谢谢!
@PeterAurelius 部分训练图是经过合成的,人脸+骷髅 让AI识别这样的组合
@XS_Rylee 好的感谢,受教了
i like it a lot , well done man









