Open jiangpeiming opened 1 month ago
I'm not sure what you mean exactly. If I understand correctly, How about adding a post-processing node?
I think the reason is the resolution in which the original model is trained, which seems to be only 256x256. At least this resolution is hardcoded in the original code for the cropper. Unfortunately the training code is not available yet.
Can the steps of synthesizing the face image and the src-image be separated to facilitate high-definition restoration in the middle? now The face clarity is a bit low after modification.we can add adetailer to repair it, the effect will be better