[Enhancement Request] FaceSwap Node: Cache faces and face_analyser model

[MoonMoon82]

Hi!

I just saw that the faces (source_face and target_face) in function swap_face are not being cached.
Also in function get_face_single the face_analyser model is loaded everytime the node is executed.

I assume it would save some time if the faces and face_analyser model is cached after the FaceSwap node is executed.
(If you agree, please be aware of caching this data in the FaceSwap node wouldn't help because it's going to be resetted immediately if any input is changed - so I suggest you to store this cache data in the LoadFaceSwapModel node.)

Kind regards

[melMass]

Thanks!
Good points! I will look into it

[melMass]

@MoonMoon82 This was auto closed by my commit, but it's indeed much faster now! I think there is still a bit of optimization possible but it's much better already!

I made this to test it out 😅

Georges Lucas as Terminator:
The player seems to no like it, but you can click here to watch it

terminator_lucas.mp4

I also updated the example file to match this update

[MoonMoon82]

@melMass This is awesome fast!! Great work!!
Just a tiny issue: antelopev2.zip and buffalo_m.zip could not been downloaded automatically. I guess these file are taken down. I needed to search the web to download them from a google drive and extract them manually.