Sadly, this will likely be an issue with your video. Most likely what you see initially is correct, but, for whatever reason, the manual tool is having difficulty keeping the video frames and the alignment information synced up.
This will most likely be fixable by re-encoding the video file.
No, it's not because of lack of RAM.
It will be some specific M1 issue, which I would be unable to advise on, not owning an M1 Mac.
The best I could recommend, is to do the Mac equivalent of this: app.php/faqpage#f1r1 and hope for the best.
Not easily, no.
Quickest and easiest way would be just run an extract on a single image, selecting all maskers (and identity), and pick up the URLs that way.
Yes. It's why we use EG/sec. Iteration is entirely dependant on batch size. 1 iteration = 1 batch through the model.
I don't follow....
I assume you are using the installer (viewtopic.php?t=20)? In which case I mean click the "Show Details" button on this:
download/file.php?id=39
Unzip it.
Faceswap does not know if the zip file is a full file or a partial download, so it has to query the server to get the file size.
If the model is unzipped, then it doesn't need to do this test.
Super resolution on video is tricky, because GANs tend to lack temoral coherence. This means that the end result is far from satisfactory
It's hard to say without images, but you could try enabling ICNR initialization which is specifically designed to avoid the checkerboard effect.
Unfortunately, for contractual reasons, I am unable to say.
Encoder doesn't output at image size. It outputs a feature map at reduced spatial dimension, but increased depth.
Ok, that is very odd.
When I'm back at my PC I will test the command to see if I can recreate the error.
In the meantime, you should be able to get extract to proceed by disabling the "Identity" option.