ZeroCool22
Banned
- Messages
- 424
- Likes
- 882
I'm starting to play with the Neuronal Networks script aka "DeepLearning", where you feed the networks with 2 sets of images/frames let's say SOURCE (VIDEO) and TARGET (the face you want to to put onto the video), then you feed the network with those images and start training till get the desired results, it's not fast (need to train for like 5 -8 hours or more), even more when face shapes are not too equal between them.
This takes some hours to TRAIN, the script use the GPU (TENSORFLOW) to do the train more fast, so it's highly recommended to have a good GPU for do the work.
All of this is just starting, need some improvements yet, but this have a lot of potential...
Here one of my first test with this:
PD: Maybe some of you heard some time ago on the news about the Gal Gadot fake porn video and how well it was done, well it was done using this method.
It have some flaws, the more noticeable is some flickering, but that depends on how hard the face/positions on the original video are, but this will be sorted out or sure, because scripts will become more and more accurated with the time. It opened a new era for FaceSwap on videos.
This takes some hours to TRAIN, the script use the GPU (TENSORFLOW) to do the train more fast, so it's highly recommended to have a good GPU for do the work.
All of this is just starting, need some improvements yet, but this have a lot of potential...
Here one of my first test with this:
It have some flaws, the more noticeable is some flickering, but that depends on how hard the face/positions on the original video are, but this will be sorted out or sure, because scripts will become more and more accurated with the time. It opened a new era for FaceSwap on videos.
Last edited: