TRANSFERRING FORGEDIT TO IMAGIC

Shiwen Zhang

Abstract: In this supplementary letter, we transfer techniques of our Forgedit to Imagic. Keeping the multi-stage fine-tuning process of Imagic, we change everthing else to our Forgedit settings. We transfer our forgetting mechanism with disentangled UNet to Imagic in order to tackle its overfitting problem during editing stage. We found that our novel techniques improves Imagic, which demonstrates that our findings in Forgedit are universal.