Resized input video (target) AFTER extraction, what to do?

[kocheko]

hello, i was trying to find some topic related with this but i couldnt so here we are.
i am a newbie with the tool so i can do the basic things, extract, clean aligment, train, convert. recently i worked on a video where the final result the face replaced was too blurry , it is ok, i dont have HD material of the face i want to swap, so i though this: "i resize the 1080 video where the original face is, to 480, this way quality will match with the souce material i have" ...ok after doing so, and try to convert, no faces were swapped and only some artifacts are shown....i know it has something to do with the aligment tool??? but how? i tried nothing and i am all out of ideas.

[torzdf]

I assume that you are using the original 1080p alignments file with the new 480p video...

If so, then yes, this won't work. You would need to generate new alignments for the 480p video.

However, an easier method would just be to feed convert the 1080p video, then use the "Output Scale" option in convert (under Frame Settings) and set it to 44 (480 / 1080 = 0.44).

This will do the swap and then resize the output.

[kocheko]

thank you very much for your help. ok i want to resize prior to swap , if i resize after swap, the quality of the swapped face will also decrease, right? so i guess i have to start from scratch with the 480p video

[torzdf]

Honestly, I doubt you could tell the difference between doing it before or doing it after, and it is certainly quicker to do it after.

If I were you, I would run a convert with the output scaling as above. Only if you then were not happy with it would I consider marking up a lower res input file.