How to fine tune Stable diffusion's Unet with LoRA

#66
by JamesWu123 - opened

I'm trying to add another information into SD, the input of SD is text, and I want to add embedding pretrained by FM into stable SD. I've done a little bit research but seems there's no same situation. My idea is to fine tune input attention, let text embedding and pretrained embedding can be in the same latent space.

Sign up or log in to comment