欢迎使用,,lora名字自己改一下,我重新改了下名
tag识别不出,请下载原图到sd反推
底模是: breakdomain_M2150 ,dalcefoPainting_3rd,anything,
embeddings模型: EasyNegative,ng_deepnegative_v1_75t,badhandv4 需要下载,,不然可能达不到效果
VAE: vae-ft-mse-840000,万能vae,,,其它vae也可以试试。色彩会有点点不一样
Sampler: DPM++ SDE Karras, Euler a,DPM++ 2M Karras,每种都有一点点不同的效果。自己测试
CFG scale: 7-9
采样: 20-40
人物占比很小。应该都是烂脸的。可以打no humans,加高权重不出人,或者开超大分辨率看看脸能不能好。不然就用插件或者其它方法修,
我一般用的是512x768,高清修复放大2倍1024x1536,或更大,不要同时用脸部修复,横图是768x512,
————————————————————————————————————————————————————————————————————————————————
Welcome to use, change the name "lora" to your own name, I have renamed it.
If the tag cannot be recognized, please download the original image to the SD card for reverse pushing.
The base model is: breakdomain_M2150, dalcefoPainting_3rd, anything.
Embeddings models: EasyNegative, ng_deepnegative_v1_75t, badhandv4. They need to be downloaded, otherwise the effect may not be achieved.
VAE: vae-ft-mse-840000, universal vae, other vae can also be tried. The colors may be slightly different.
Sampler: DPM++ SDE Karras, Euler a, DPM++ 2M Karras. Each has a slightly different effect. Please test it yourself.
CFG scale: 7-9
Sampling: 20-40
The proportion of people is very small. They should all have bad faces. You can use "no humans" and increase the weight to exclude people, or try to view the face with a very high resolution. Otherwise, you can use plugins or other methods to repair them. I usually use 512x768, and high-definition restoration can be enlarged by 2 times to 1024x1536 or larger, but do not use facial repair at the same time. The horizontal image is 768x512.