@4044 I think I found the issue. I was using DPM++ 2M Karras as a sampler but the ones in yours and others examples have "TEST" at the end. So that must be a different sampler (newer version I guess?). When I switched to Euler A, the other sampler you're using, everything rendered with more detail and doesn't look so out of focus, then when I did my upscaling I got the higher def results I normally see.

@Snipes hey there! I'm also having a bit of trouble getting faces to look high res. Can you explain this step here? inpaint the face-->only masked-->set to 512x512-->Denoising strength:0.2~0.5

I'm new to stable diffusion and I just want to understand what this means. More specifically, what is inpainting?

@Hiatus I don't use impainting much, never had much success with it. Just click "Restore Faces" and "Hi Res Fix" in txt2img, upscale by two. Then send that result to img2img and upscale more if your want. Faces always turn out great for me that way. Make sure to deselect "Restore Faces" in img2img you don't need to do it a second time and it usually messes it up if you do.

@4044 这是我用过最好的模型,应该说是出来的图最大限度符合脑海中的想法,准确率高,同时各种风格适应性极强.希望作者再原模型上能够进一步加强,如东方古代神话人物.

That's exactly what I was going to say! It's super impressive.

应该是自带的吧。。就是ui界面的采样方法（sampler） 下拉选单里的。 默认是 Euler a

你自己做的图，要啥版权？

版权小鬼来什么C站？

@aose C站是版权法外之地吗

属于是连AI绘画的原理都没明白

@syh9798 你都翻墙了，你还谈法律？谈法去libulibu，出来了就别搞笑了。

设置小尺寸。然后勾选Hires.fix （两步生成），放大倍数1.5-2，重绘选择0.5左右

@wontonaaa675 生成的图不够精细，而且没有原图那种干净感，请问您用哪个放大算法？

@orange0329 可以用高清修复，有关高清画质的提示词，或者用图生图2倍放大

那倒未必

确实，随机性太大了，整体上来说可能是提示词污染的比较厉害，风格固定比较差。