what's the training difference between with and without "light"
#22
by
sutongtong
- opened
According to description, checkpoint with "_light" is more compatible with text prompt. What's the difference during training?
for light version, we random crop image as condition in the training stage