XUANmix版本v2.0 (ID: 470253)

Filename: XUANmix__v2.0.ckpt

This checkpoint is not mine! I got it from this: https://tusi.cn/models/605939393502388131
That website used to allow downloading for free (I downloaded it at that time), but now it requires an account (which asks for a CN phone number or WeChat to register).

So I uploaded it here for convenient usage. If you are the original author and don't want it to be here, please let me know and I'll take it down.
(Idk whether it's trained or merged checkpoint. There's no info about checkpoints that are being merged so I assume that it's a trained checkpoint).

No baked VAE so you need to download one. I'm using "blessed2.vae.safetensors" and it suits my liking. You can also use "vae-ft-mse-840000-ema-pruned.safetensors" but I think the saturation would be too high.

The first two sample images are from the original post. Using A1111's SD, I can generate the similar images by changing ENSD in the settings to 31337, and using "vae-ft-mse-840000-ema-pruned.safetensors" for VAE.
Other samples are generated with default ENSD, "blessed2.vae.safetensors" for VAE, and with Adetailer to enhance the face.

Settings taken from samples of original post:

• Sampling Method: Euler a

• Sampling Steps: 25

• CFG Scale: 7

• Clip Skip: 2

• ENSD (Eta noise seed delta): 31337

• Size: 512x768

• Upscaler: R-ESRGAN 4x+

• Hires Steps: 30

• Denoising Strength: 0.2

Original description from source:
（建议开高清修复使用）生成半身像的效果比较好，全身的因为我显卡不太行图片不能太大，效果一般般，分辨率大一点或者搭配人物Lora的话效果可能好点,

描述:

训练词语: