Faceswap Forum

Also note that the default BS of 64 is really only usable for Original on 8gb. For Villain you'll need to adjust that down to your system.

Faceswap cannot do this sort of operation, it's meant for moving one face onto another not to change one face's expression/pose to match an input. To do that, you'll want to examine tasks like re-animation (I suggest the Ganimation implemenation, though there is no trained model available).

Alignments are really not that important, they're used to get the general face and to make one for of mask, but we're already starting to support much more advanced masks.

The best quality will come from manual composition, not manually tweaking close but not quite perfect alignments.

Please try manually selecting the alignment file in the Alignments box and try that way. It may be missing the alignments file when it looks for it.

In the conversion process you include the alignments file for the Video, NOT the face. So if you have the Alice in Wonderland with the faces you want selected, then you'll include that alignment file and no others.

We currently don't have an ETA on the conversion guide as Conversion is going through a rework and so the guide is waiting for the end of that process. Conversion though is relatively simple. The most basic operation would be to put (in order on the GUI) the Input folder or video, the output folder,...

The problem you're seeing is because the driver is not the same as the cuda version. You need to upgrade the graphics driver on your instance to match your cuda version.

See
https://docs.nvidia.com/deploy/cuda-com ... index.html
For a graph matching cuda version to required driver versions.

There is not enough information here to diagnose. Either post the full log and your command or the faceswap.log file from the faceswap folder.

Masks have been available in Faceswap in various ways for a long time. We're just moving into a new type of automatically generated mask than the previous versions. The new one should be more accurate and flexible moving forward.

MemoryError Exception occured trying to retrieve sysinfo: [WinError 1455] The paging file is too small for this operation to complete The problem is that you don't have enough RAM for the operation. We're currently tracking some reports of increased RAM usage on some situations and if you could pos...

Loss is a poor measure, don't even bother using that as a guideline (especially if you're using a mask since that gets added to the loss). Use the quality of the faces, and if you aren't getting good enough results, then you need to add more data. I'd leave those images in, especially if they're in ...

You only need to merge if you have more than one alignments file that you want to combine. If you've only got the one, you can skip that section. If you have two that you need to join, you just put both files in the alignments textbox with a space separating them.

There is a manual tool, however, you don't actually need to adjust the mouth (and we have no method to adjust them individually). We only use the average landmark in order to get general alignments so slightly off mouth doesn't matter. The sides of the face can matter for masking if you're not using...

Use the manual tool only in the last resort. The sort tool will be much faster and more reliable. But if you really need the manual tool, make sure you're in "edit" mode (use the "m" key).

You need to clean your alignments file to make sure that only the faces you want swapped are included. Please see the Extract guide viewtopic.php?f=5&t=27 on doing that.

You can use relative paths and FS will use them if you provide them. The thing is that the selection tool wont easily use relative paths, so if you use the GUI to select a path it'll provide absolute path. But you can easily provide a relative path by just putting the parts after the FS folder (whic...

A quote from Niko about the stuff he was using. nikopueringer 07/14/2019 Answering a couple of the questions from further up; We used both FS and DFL. In FS we used the villain model, with 100% coverage and extended mask. I first pretrained the villain model for 50k iterations on random faces, then ...

Does this happen multiple times or was it a one-off? If it's hapenning all the time, can you provide the video that you're using (Post it somewhere and PM the link is fine) so we can see what's happening?

More specifically, you can feel free to change the batch size as you train. Lowering the BS to ~64 near the end of training makes sense if you want to see the differences of a smaller BS. It wont hurt to try different values if you want to see how it affects your results.

This looks like a problem with "on the fly" conversion. Try providing an alignments file from the extract step and see if the problem resolves.