You are viewing a single comment's thread.

view the rest of the comments →

0
2

[–] ambivalenthoe 0 points 2 points (+2|-0) ago 

It's strange because the best deepfake ever is the Gal Gadot one and the model looks nearly identical but some of the new fakes ( like the new emma roberts one) you can see in the screenshot that the two look nothing alike. And yet it looks very convincing.

There seems to be something very technical going on when it overlays the original face like that. In general for newbies though, the big thing is that they look as close as possible in the scene being used. In terms of hair and (mainly) head shape but if you're not as skilled at training; the face shape needs to be a doppelganger as well. That's just what I've noticed from observation.

Also: training too long can result in overtraining, causing a reversal back to the original model. Training way longer does not mean it will be better.

0
0

[–] Robjones1234 [S] ago 

Your observations remind me of a theory I have about deepfakes ... It works best as digital makeup. In essence, you are smoothing over one girls features with another. Your layering a different nose cheeks eyes etc over a base face. I for one am shocked that similar results could not be achieved in Photoshop as I know they have a digital makeup type feature in the software. But then again, deepfake is way more intricate than a head or face swap (which is still a convincing way of taking in its own right)