lowpoly_nijimix

欢迎使用，这是一个低多边形风格的模型，

Welcome! This is a low-poly style model

embeddings模型：     EasyNegative，ng_deepnegative_v1_75t，badhandv4

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

出全身要用大尺寸图，否则脸崩，我用的是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,奇怪的尺寸会出畸形多手，多头等，不懂最好用这个尺寸，，

高清修复好像是latent低重绘权重会好点，我的高重绘烂手概率会很大，我的显卡可能有问题，低重绘会模糊，请自己测试。

____________________________________________________________________________________________________________________________________________________________________________________________________

embedding models：EasyNegative, ng_deepnegative_v1_75t, and badhandv4.

VAE models ： vae-ft-mse-840000,, and other VAE models can also be tried. The colors may vary slightly.

sampler ：DPM++ SDE Karras, Euler a, and DPM++ 2M Karras. Each has slightly different effects, so you can test them out for yourself.

CFG scale： 7-9,

sampling : 20-40.

If you want to generate full-body images, it's best to use a large image size, such as 512x768, and then upscale it by 2x to 1024x1536 or larger. Avoid using face restoration at the same time, and use a horizontal image size of 768x512 to avoid distortions, extra hands, or multiple heads.

For high-quality restoration, it seems that using a low-latent redraw weight will produce better results. There may be a higher chance of poor results if using a high-latent redraw weight, so it's best to test it out for yourself.