v2: 2023-02-20 I erased the roughly made picture and put in the painstakingly made picture. Do img2img SD upscaling. img2img SD upscale method: scale 20-25, denoising 0.2-0.3 After selecting SD Upscale at the bottom, tile overlap 64, scale factor2 caution! Sampler must be DPM++SDE karras. If you want to make it more realistic, use DPM++ SDE Karras. If you want a clean face, use eluer a. Have fun using it. I used vae kl f8 anime for the example meme, but I recommend using 840000. clip skip 2 https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.ckpt
https://huggingface.co/AIARTCHAN/aichan_blend/tree/main/vae
Apply VAE.
You will get better color results.
v1: I don't like the model with too many prompts and forcing you to use LoRa. So it was created. Let's extract high-quality images with 6-word prompts! Check the sampler, scale and step in the example image. Looking at the example image, there is a repeating prompt. VAE: vae-ft-mse-840000-ema-pruned Skip Clips: 2 It was a very simple test, so I think there will be many better results. I didn't even do upscale testing. Have fun while testing! !There is one downside. It's fine for dark eyes, but for other eye colors, the pupils often collapse without upscaling.
|