stable-diffusion
stable-diffusion copied to clipboard
What is 10% dropping of the text-conditioning?
Hello, the card for https://huggingface.co/CompVis/stable-diffusion-v1-4 says it does 10% drop of text-conditioning?
What does that mean?
It means dropping 10% of the captions during training (namely 10% of the images are trained unconditionally). However, I am not sure about which part in the code does this. Can someone point out to me?