Arrrgh. As usual, I have hit a stumbling block with Wav2Lip. The Google Collab links don't install but don't give me an error message on which I can work (the python installation just hangs - all the more annoying since I know I have the right version of Python already installed but possibly not in the right folder and it won't complete the next step without completing this one.
So I figured I would try a manual install on my laptop since my GPU is powerful enough to cope. Github can be a bit impenetrable if you have no training in coding. They rarely spell out the process in laymans terms (I think because they want to discourage the average member of the public from abusing the technology. The manual install just says 'clone the repository' but the commands listed are not the commands you need to do that. It then says save the pre-trained models and gives you a file name but doesn't tell you WHERE to save them. You watch the help videos and they say, 'this took 10 minutes' but after 4 hours I am barely at stage 2. I give up.
EDIT: Obviously, I didn't give up. I managed to work out the collab issue was a Google docs path error because they'd taken the space out of My Drive. Sigh. It still isn't working in that it's just hanging without producing anything but I at least I don't feel defeated by stupid syntax.
I still have no idea how to manually install it...
EDIT 2: Ok - eventually got it to run but only with really low resolution. One layer of Deepfake later and the results are in. I tried to mask everything but her face using layers. The lipsync is improved. i will have to see if the lower resolution is that noticeable on a bigger screen.
I did run it for her other scene but looking at the resolution on this one, I can't see that it's going to look as good on more of a closeup. I tried to run it Kirk's space suit as well so the lipsync for "I have him in sight" syncs better with, "Just like old times." but the image was too large. I ran this image by cropping out Kirk, so it might work if I crop everything but his face. Worth a try next time I have a bit of time.
Edited: Swapped for a video that shows how I managed the process.
So I figured I would try a manual install on my laptop since my GPU is powerful enough to cope. Github can be a bit impenetrable if you have no training in coding. They rarely spell out the process in laymans terms (I think because they want to discourage the average member of the public from abusing the technology. The manual install just says 'clone the repository' but the commands listed are not the commands you need to do that. It then says save the pre-trained models and gives you a file name but doesn't tell you WHERE to save them. You watch the help videos and they say, 'this took 10 minutes' but after 4 hours I am barely at stage 2. I give up.
EDIT: Obviously, I didn't give up. I managed to work out the collab issue was a Google docs path error because they'd taken the space out of My Drive. Sigh. It still isn't working in that it's just hanging without producing anything but I at least I don't feel defeated by stupid syntax.
I still have no idea how to manually install it...
EDIT 2: Ok - eventually got it to run but only with really low resolution. One layer of Deepfake later and the results are in. I tried to mask everything but her face using layers. The lipsync is improved. i will have to see if the lower resolution is that noticeable on a bigger screen.
I did run it for her other scene but looking at the resolution on this one, I can't see that it's going to look as good on more of a closeup. I tried to run it Kirk's space suit as well so the lipsync for "I have him in sight" syncs better with, "Just like old times." but the image was too large. I ran this image by cropping out Kirk, so it might work if I crop everything but his face. Worth a try next time I have a bit of time.
Edited: Swapped for a video that shows how I managed the process.
Last edited: