The English introduction follows the Chinese introduction, you can click "show more" to view it.

如果你想生成壁纸级的图片，请一定要按照我说的做。

所有图片还是建议使用高分重绘生成，你会发现质量比直接生成的高分辨率图片高太多。

高分重绘参考图：

CFG Scale建议3或2.5

关于lora的权重，换衣服的话0.4左右，不换衣服随意，但不要超过0.7

底模的话，像什么AbyssOrangeMix、FaceBombMix基本都可以，不过强烈推荐CetusMix，我心目中的神级模型

介绍一个功能：SD Upscale。SD Upscale具有提高分辨率和补充细节的双重作用，经过SD Upscale处理，你的图片会非常清晰且富有细节。SD Upscale在图生图的脚本里开启

这里是我的SD Upscale放大参数， 不要将重绘幅度设在0.4以上, 否则会破坏画面结构，提示词和负面提示词直接复制粘原图的

如果你有任何问题，可以在评论区告诉我，我会尽力解答。

If you want to get images that look as good as wallpapers, PLZ do what I say.

I highly recommend using hires. fix to generate images, you'll find it's much better than directly generating images in high resolution.

This pic is for reference:

CFG Scale~3 or 2.5

Setting this lora's weight around 0.4, you'll find it's not only easier to change her clothes, but also still looks like Ganyu.

you can give more weights to clothes and postures if it's difficult to change them(above 1, even above 1.5 is still ok).

Some full-body images have been processed with SD Upscale, it's a very useful script which can make faces in full-body imgs much richer in detail, you can find it in img2img.

This pic is for reference if you want to use SD Upscale, Do not set denoising above 0.4 or the image will collapse. Prompts and negative prompts in img2img must be same as those in txt2img

SD Upscale has the dual function of increasing resolution and adding details. After being processed by SD Upscale, your images will be very clear and rich in detail.

If you have any questions, PLZ tell me in the comments section, I'll try my best to help you.