Welcome to my personal favorite checkpoint. It's a checkpoint aimed at being anime centered with a painting/stylized theme. It's very flexible, and works great with most loras. This model does both SFW and NSFW images.

Recommended settings:

VAE:

First of all you should be using a VAE which you manually set in the settings. Download either:

https://huggingface.co/stabilityai/sd-vae-ft-mse-original/tree/main

or

https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt (I use this one)

Place this file in your stable diffusion directory -> models -> VAE

Then on the webui go settings -> stable diffusion -> SD VAE and choose from the drop-down menu.

If you do not use a VAE, all your images will come out washed out and grayish.

Generation settings:

Resolution: 512x512 or 512x768 or 768x512

Steps: 30-40

Sampling method: Whatever you like best. My personal favorite is Euler a, but i also use DPM++ 2M Karras and DPM++ 2M SDE Karras

Hires.fix:

Upscaler: 4x-UltraSharp or 4x_fatal_anime_500000_G can be found in: https://upscale.wiki/wiki/Model_Database

Hires steps: 10

Denoising Strength: 0.3

Upscale by: Whatever resolution you want. I usually go 1.75

描述:

A more painting/grainy style checkpoint. Works better with different styles when prompting for them.

训练词语: