Faceswap Forum

I am very new to the forum but having fun learning. I'm trying a "lightweight" swap to practice. I had a disaster on my first try, and I'm on my second.

I'm reading the importance of source data, and I want to manually adjust the masks and landmarks to be accurate. My a and b data have tens of thousands of frames extracted.

New-guy question: Are users of the app manually adjusting tens of thousands of frames individually? It seems like you cannot "apply" the change across the set. You must change the alignment for one frame, then move on the second frame, then move to the third, etc.

Am I missing something? Are we generally meant to leave these alignments alone unless there is a giant error with the alignment? Or is it better to fine-tune them each?

Generally you'll only want to fine-tune data that you're using for actual conversion. It's not worth fine-tuning thousands of files for training. Better to just delete bad images rather than tuning them.

Thank you!! This leads me to a follow-up question. Let's say I'm training perfectly (I'm definitely not, but just for argument's sake). So I have a good model.

At that point, would I be able to convert a 50-frame video and fine-tune every single frame for the best result? I've been using these massive videos so far for the conversion and I am starting to think that might have been a little ambitious to begin with.

I guess my actual question is: Would it be 'easier' or 'safer' for a newbie like me to try to spend a long time on training, but convert a very very short video?

Short scenes let you focus your efforts. Even if you're doing large amounts of scenes, I recommend converting one at a time. It lets you tweak settings and efforts to get the best results for each situation.