Search found 2617 matches

by torzdf
Sat Mar 02, 2024 1:13 pm
Forum: Training Discussion
Topic: Newbie - should I continue with this data/training
Replies: 1
Views: 91

Re: Newbie - should I continue with this data/training

Villain is a notoriously slow model to train. so you may want to switch to something like dfaker to get quicker results (at slightly lower quality). Note that neither of these models will be good for close-up shots as they output at 128px. Bigger models require more VRAM and more time to train thoug...
by torzdf
Sat Mar 02, 2024 1:09 pm
Forum: Extract Support
Topic: Crash during Extraction
Replies: 2
Views: 90

Re: Crash during Extraction

This looks like an edge case which may or may not be related to AMD. If possible, could you provide me with the source you are truing to extract from (DM is fine), I can see if I can recreate the bug in order to fix it.

by torzdf
Sat Mar 02, 2024 1:06 pm
Forum: Extract Discussion
Topic: Manual Tool and faces filters
Replies: 3
Views: 132

Re: Manual Tool and faces filters

Honestly, I can't remember off the top of my head.... Again, a workaround, if you use Generate instead of Extract , you can copy the command, open up an Anaconda prompt do: Start > Anaconda prompt conda activate faceswap Then paste the command there and run. There is no extract/convert preview in th...
by torzdf
Fri Mar 01, 2024 11:23 am
Forum: Extract Discussion
Topic: Manual Tool and faces filters
Replies: 3
Views: 132

Re: Manual Tool and faces filters

> How difficult would it be to add some face filter more to manual tool? I'd really like to have "only 1 face", "2 faces" and "over 2 faces". Filters existing doesn't help the way I like because "has faces" shows every frame with faces despite the amount of th...
by torzdf
Fri Mar 01, 2024 11:05 am
Forum: Convert Discussion
Topic: Conversion time increased suddenly
Replies: 2
Views: 133

Re: Conversion time increased suddenly

Nothing has changed in conversion for a while... The most likely causes is that you are using an h265 video as the source (when you have previously used h264 sources in the past). h265 takes longer to decode.

Also, if you are training a more complex model, this will also take longer to convert.

by torzdf
Fri Mar 01, 2024 10:59 am
Forum: Training Discussion
Topic: Suggestion for a best model for me and a swap issue
Replies: 1
Views: 78

Re: Suggestion for a best model for me and a swap issue

6GB will be a challenge for high quality results. In terms of available models, then dfaker or dfl-sae are likely to be your best candidates. Once you are more comfortable with the process, then you can look to load one of these in Phaze-A (from a preset) and see if you can tweak it at all to get mo...
by torzdf
Mon Feb 26, 2024 10:07 am
Forum: Extract Discussion
Topic: Faceswap face extraction
Replies: 0
Views: 312

Re: Faceswap face extraction

> So the face extraction just extracts the faces from the frames at a lower res, This is not strictly correct. It extracts, and aligns. faces at a uniform resolution to be able to train a model. This may be at a higher resolution, lower resolution or the same resolution as the original image. > is t...
by torzdf
Tue Feb 20, 2024 7:28 pm
Forum: Training Discussion
Topic: About Mask while training and after that
Replies: 1
Views: 581

Re: About Mask while training and after that

loss) there is a check box for choosing mask https://imgur.com/a/PcgU2g5 . From my understanding the selected mask will focus masked face area while Training. Right ? And when we convert there are many mask that we can choose (mask that included in alignment). Yes, the mask you select in settings i...
by torzdf
Tue Feb 20, 2024 1:56 am
Forum: Extract Support
Topic: Error generating mask due to invalid tensor format fp32
Replies: 1
Views: 593

Re: Error generating mask due to invalid tensor format fp32

This is a bug in Tensorflow-Metal (Apple's plugin layer) so is upstream from us. Ultimately "Invalid tensor format: "fp32" doesn't make a lot of sense. Nearly all tenors are fp32! If you are getting full output, then do not worry about it too much. If not, then there may be bigger iss...
by torzdf
Wed Feb 14, 2024 1:54 pm
Forum: Training Discussion
Topic: A Short Hello with some questions
Replies: 3
Views: 569

Re: A Short Hello with some questions

If I understand you correctly, it comes down to this (fictional/theoretical): If I want to "smuggle" myself into an Indiana Jones like Harrison Ford, I shouldn't extract the entire film, but rather mainly passages with his face from all sorts of movies, so that Harrison's face becomes bes...
by torzdf
Tue Feb 13, 2024 12:34 pm
Forum: Installation Support
Topic: Specify a GPU for SwapFace use?
Replies: 1
Views: 707

Re: Specify a GPU for SwapFace use?

Yes. The command line option -X (or the gui option exclude GPUs in the Global section of each page) lets you exclude GPUs for use in Faceswap. We have .dockerfiles in our repo. You will need to mount some volumes for data/persisting config changes: https://github.com/deepfakes/faceswap/blob/master/I...
by torzdf
Thu Feb 08, 2024 12:24 pm
Forum: General Discussion
Topic: Putting Words in Someone's mouth (literally)
Replies: 1
Views: 827

Re: Putting Words in Someone's mouth (literally)

You can't do that with Faceswap, no. The video A drives the swapped face, so you would need a video of your professor saying those words already, which would undermine the whole point. To do what you want, you would need an "actor" to say the words you want and then swap the faces. Look in...
by torzdf
Tue Feb 06, 2024 12:30 pm
Forum: Training Support
Topic: OSError: Unable to synchronously open file (bad object header version number)
Replies: 2
Views: 1096

Re: OSError: Unable to synchronously open file (bad object header version number)

Your model file is corrupted.

You can either use the model tool to restore from backup, or you can delete the current model folder and rename the last snapshot that was taken.

by torzdf
Mon Feb 05, 2024 12:34 pm
Forum: Extract Discussion
Topic: Extraction question about object on the face & weird too-wide jaw line alignment
Replies: 1
Views: 961

Re: Extraction question about object on the face & weird too-wide jaw line alignment

Do you include picture(s) of face that some part of faces are hidden by objects (like book, leaves, tree, or wall...) to training data? If they are masked correctly, I do include them, yes. As I mentioned above, I used Bisenet-Fp masker during extraction. Is Bisenet-Fp masker landmarks based mask, ...
by torzdf
Mon Feb 05, 2024 12:32 pm
Forum: General Discussion
Topic: New faker general questions
Replies: 6
Views: 1127

Re: New faker general questions

Honestly, I have no idea. Windows 7 is well out of support now, so it may or may not work. Not been tested.

by torzdf
Sun Feb 04, 2024 2:00 pm
Forum: General Discussion
Topic: New faker general questions
Replies: 6
Views: 1127

Re: New faker general questions

You can do that, yes. However, if you are happy with the final model, then there is no need to keep the snapshots (unless you wish to specifically rollback to a particular point in the model's history) re: images, that is correct. You can feed the model any faces you wish. No need to keep the origin...
by torzdf
Sun Feb 04, 2024 1:22 pm
Forum: Training Support
Topic: Face correction
Replies: 1
Views: 919

Re: Face correction

This is entirely a data issue. I assume you are saying that the A->B swap looks bad? The reason this occurs is that you have faces tilted at more extreme angles in the A (source video), but do not have sufficient coverage of that data on the B side. When the model does the swap, it struggles to crea...
by torzdf
Sun Feb 04, 2024 1:20 pm
Forum: General Discussion
Topic: New faker general questions
Replies: 6
Views: 1127

Re: New faker general questions

So, just started for the first time a few days ago, read a ton of guides and posts, did some searches and answered just about all my early questions and have hit a snag on a topic. Welcome to Faceswap and the forum :) I know the model training folder is the training data set which is a must to back...
by torzdf
Fri Feb 02, 2024 3:54 pm
Forum: Training Support
Topic: Error starting training: TypeError: 'NoneType' object is not subscriptable
Replies: 5
Views: 1087

Re: Error starting training: TypeError: 'NoneType' object is not subscriptable

I meant, specifically, what is the contents of those 2 folders. What is happening here is that the process is trying to read the Faceswap meta information from the images in these folders. It is unable to. The only reason this would happen is if you had images in that folder that were not generated ...