I can't guarantee this won't break everything, but you can try:
Code: Select all
conda activate faceswap
conda install "numpy>=1.24.0"It will either work, or you will end up with a lot of errors telling you about conflicting issues. Hopefully the former.
Doing a bit of googling around, this issue appears to be driver related. As it has only happened once for you, I would not worry too much. If it happens more consistently, look to update your GPU drivers.
That's good to hear. As you've probably gathered, I haven't had a chance to investigate the data you sent me, but seeing as it all appears to be working, I will shelve that for now.
That is correct, as it fundamentally alters the target of the model's training, so mask type is locked to model.
You cannot change masks for existing models. You will need to start a new model to change the mask type.
The mask tool can be used to output previews of the masks stored in the alignments file/png header.
Without seeing a screengrab of the training preview Window, it is hard to say why you may be having issues.
No need to retrain. You just need to generate an alignments file for the target conversion video. See here for more detail: viewtopic.php?t=27#align
"Possibly" is the best I can offer.
Intel/Nvidia Macs haven't been directly supported for a while now, although the libs remain in Faceswap meaning that they should still work. It's not been used/tested for a while though.
This is an OOM (Out of Memory) error. See here: app.php/faqpage#f3r9
Your GPU Driver is very old, so you should look to update.
Link posting is limited for moderation reasons. However, I have changed the setting to allow the post into the moderation queue, rather than denying links outright.
That is a hairline and/or data variety issue. If it's the former then you can read more here:
viewtopic.php?p=7684
If its the latter, then you need more varied data (example of highly varied data: app.php/faqpage#f3r0)
You can add and remove training images at will. Just make sure to stop training before making any changes, and then resume after.
Also: app.php/faqpage#f3r4
It looks like the model file has corrupted on download.
Go into your faceswap folder and into the .fs_cache folder and delete the file .h5 model files in that folder. This will cause the process to re-download them.
Looks like model file corruption for some reason.
See here to restore from backup/snapshot and get back on your way:
viewtopic.php?p=8361#p8361
Honestly, no idea. they should work the same (as they have the same VRAM), but some kind of overhead appears to be applying to the 3070 card. You can check your VRAM usage prior to training to see if the 2nd system has more VRAM allocated by the OS.