Faceswap Forum

For some reason, I've noticed that the effective area shown in the preview for training doesn't touch the edges of the pics either for source or destination. This is causing some weirdness like when the face goes out of frame for a second, there's a thin strip along the edges where the conversion doesn't touch. So I see the original face sliding by until it's out of the "deadzone".

Unfortunately I think I would need to see an image to really understand what you mean.

I'm still not quite sure what the issue is here?

If you are talking about the 'inner' box, then just increase the coverage amount in training settings.

The issue is that the red masking area doesn't extend to the full width and height of the frame so there's a small sliver around the edges of the face that don't get the replacement face. If this were just a training thing, so be it, but it affects the final converted video as well. Most importantly, it didn't used to do this and I don't recall changing any settings. But you're saying that it might stop if I push it to 100% coverage?

Yes. Up the coverage 100% will use the full image you see above. Anything less than 100% will use the smaller box that is shown.