kohya-ss

/

controlnet-lllite

like 121

So when I'm using the blur model, I use the blurry picture for the conditional image in lllite, but then should I sample to a new empty latent or should I encode the blurry image and then denoise it?