two tricks to improve results.

[yangchen8710]

adjust the new face's average color as same as the old face's

use the old face's edge to apply a smooth mask to new one

[Jack29913]

Looks good. Can this be done automatically in the pipeline?

[yangchen8710]

@Apollo122
This is a result of automatic.
Already done the code.

[Jack29913]

@yangchen8710
Great. You can make a pull request if it's ready

[Ganonmaster]

These changes have been merged. Share your results in the faceswap-playground!

[deepfakes]

@yangchen8710 many thanks for this amesome contrib!

It's amazing how you did it with such simple approach! I added you as collab on the 2 side projects (playground and models) so you can freely make things move on if you want. On the main project, I let @Ganonmaster handle things as he does this like a boss ;-) But feel free to push other things...

One question though. Does this cover the issues #8 and #16 about masking and alignment ? What should we do with them, close them or continue ?

[Jack29913]

I think they should remain being open. So that other people may come and improve those. #8 may use an edit though as it's not the current case anymore

[Clorr]

Hmm, we can't let issues opened just for improvement. We already have the faceswap-model for that.

The question is then:

• should we add seamlessclone?
• how is this code on the alignment? Is it worth continuing on Adding 1adrianb/face-alignment to project #16 ? (Considering pytorch is a heavy dependency)

[Jack29913]

Now that you put it like that agreed.
As for the questions,
yes seamlessclone should be added;
In my view most significant problem with pytorch is that it's not working for Windows. But there are people working on it. pytorch/pytorch#494
So in time we may get an update