First, it was experiment. I made (write of you need it) a script of downloading images form gettyimages AND RECOVER from watermarks, than I downloaded 6000 images if SLIPKNOT band. Next, sort band members and mask types. I decided to train each of members. I thought it's not trained, but result is AMAZING. Alse see, I used only 2 layers in lora, so it very small size. Trained about 10h on my 16gb 4060ti.

Next, if you are intested, will train other masks, write in comments...

I started with "classical" masks, if it will be much downloads and comments add more masks.

How to generate better and fast image in Comfyui:

0) use "hyper lora 8 steps" and this lora

1) 1st pass. 6 steps with noise enjection after 3 steps

2) 2-nd pass 6 steps img2img (from latent from 1-st pass) with 0.3 noise

And I need you put likes - thumbs-up on top right 👍!