Search found 596 matches

by bryanlyon
Mon Apr 26, 2021 8:01 pm
Forum: Training Support
Topic: Cant use images smaller than 256x256, just stops with error 3221225620
Replies: 3
Views: 73

Re: Cant use images smaller than 256x256, just stops with error 3221225620

Errors with codes like that are all coming from upstream code, not FS itself (though may be caused by FaceSwap). In this case, the error is a divide by zero error. However, I think you may be looking at the resolution incorrectly. Why are you giving it small images? If it's extracted, I HIGHLY recom...
by bryanlyon
Fri Apr 02, 2021 5:23 pm
Forum: Training Support
Topic: No data to graph for selected session after an electricity blackout
Replies: 1
Views: 483

Re: No data to graph for selected session after an electricity blackout

Looks like the power outage happened while it was saving. Luckily we create backups and you can restore it simply using the restore tool (Tools > Restore in the GUI, or python tools.py restore -h from the cli).

by bryanlyon
Fri Apr 02, 2021 1:34 am
Forum: General Discussion
Topic: April Fools -- Big news, FaceKoin announcement!
Replies: 16
Views: 2444

Re: Big news for April 1st, FaceKoin announcement!

Thank you all for participating in our April Fools day announcement. We hope you all had fun.

by bryanlyon
Fri Apr 02, 2021 1:19 am
Forum: Cloud Discussion
Topic: Any experience with preemptible resources and/or FUSE?
Replies: 4
Views: 829

Re: Any experience with preemptible resources and/or FUSE?

You could set an inotify for any changes to the .h5 and copy the files over the network if a change is detected. Just don't overwrite on the remote until you can verify the copy was good and you'll be fine. The reason that training isn't great on a FUSE is the training data needs to be loaded/proces...
by bryanlyon
Thu Apr 01, 2021 6:27 pm
Forum: General Discussion
Topic: April Fools -- Big news, FaceKoin announcement!
Replies: 16
Views: 2444

Re: Big news for April, FaceKoin announcement!

We'll release more details on it's operation tomorrow, April 2nd.

by bryanlyon
Thu Apr 01, 2021 5:39 am
Forum: General Discussion
Topic: April Fools -- Big news, FaceKoin announcement!
Replies: 16
Views: 2444

Re: Big news, FaceKoin announcement!

When turned off, it'll still mine 15% of the time. This is done to cover dev costs. But you can make that back if you allow FaceSwap devs to have 50% commission on the NFT instead.

by bryanlyon
Wed Mar 31, 2021 11:31 pm
Forum: General Discussion
Topic: April Fools -- Big news, FaceKoin announcement!
Replies: 16
Views: 2444

April Fools -- Big news, FaceKoin announcement!

Announcing today: FaceKoin. We are pleased to announce that early access enrolment into FaceKoin (Ticker code FaKo) will be opening very soon. (within a few days of today 4/1) What is FaceKoin? FaceKoin is a digital currency which incentivises distributed, multi-node AI training by rewarding “mining...
by bryanlyon
Tue Mar 30, 2021 2:57 pm
Forum: Training Discussion
Topic: training with profile faces
Replies: 5
Views: 1333

Re: training with profile faces

No compositing is the process of combining two images into one. Compositing in my example would mean using video editting software like After Effects to combine the images. You would probably need to do it frame-by-frame in order to get the best results.

by bryanlyon
Tue Mar 30, 2021 2:54 pm
Forum: Training Discussion
Topic: Would a B&W data set (on both sides) train faster than color?
Replies: 6
Views: 1093

Re: Would a B&W data set (on both sides) train faster than color?

Unfortunately that's an intuitive guess that doesn't prove out in the real world. In fact, having multiple colors can actually give the model a better chance of finding details to grab onto and get results faster. In the end, B&W is a poor optimization for speed. Though if you have old data only...
by bryanlyon
Tue Mar 30, 2021 2:50 pm
Forum: Cloud Discussion
Topic: Any experience with preemptible resources and/or FUSE?
Replies: 4
Views: 829

Re: Any experience with preemptible resources and/or FUSE?

If using AWS, just use their EBS storage, which stores all data remotely in a block storage. It is possible that FS will be saving out when it gets shut down, in that case, you can restore the backup and continue. GCE and Azure offer similar options for their premptable instances.

by bryanlyon
Mon Mar 29, 2021 3:38 pm
Forum: Installation Support
Topic: [Guide] Using Faceswap on Nvidia RTX 30xx cards
Replies: 87
Views: 17774

Re: Current hardware recco + 3090 advice needed

You have a 3090. If you want it to work with that 3090 you must follow that thread.

by bryanlyon
Tue Mar 23, 2021 3:33 pm
Forum: Training Discussion
Topic: training with profile faces
Replies: 5
Views: 1333

Re: training with profile faces

Profiles are always tough and often require postprocessing to get the best result. The problem is that you're blurring the edges and the edges in that case has a new nose sticking out past the old one. That's getting blurred with the background. There is no automatic solution for something like that...
by bryanlyon
Mon Mar 22, 2021 9:27 pm
Forum: Convert Support
Topic: Looks good during training but not after converting.
Replies: 6
Views: 1067

Re: Looks good during training but not after converting.

Try the preview tool, it helps you see the results of your convert settings on the video itself.

by bryanlyon
Mon Mar 22, 2021 6:00 pm
Forum: Training Discussion
Topic: RealFace or Dlight when training for 128px output?
Replies: 7
Views: 1706

Re: RealFace or Dlight when training for 128px output?

I'd leave it at 64x. Maybe 128x at most.

by bryanlyon
Mon Mar 22, 2021 5:57 pm
Forum: Extract Discussion
Topic: Using Manual Tool Properly
Replies: 3
Views: 1433

Re: Using Manual Tool Properly

For training, I recommend just deleting the poor profiles. To be clear: Poor profiles is when the profile isn't centered and facing the right direction. For convert, you need every face that you want to swap. However, it might be best to choose a better clip to convert if there are a lot of profiles.
by bryanlyon
Mon Mar 22, 2021 5:01 pm
Forum: General Discussion
Topic: Advice to maximize results for low VRAM cards
Replies: 4
Views: 1030

Re: Advice to maximize results for low VRAM cards

Lowmem on Original wont provide the same results as Original without. Lightweight is even more memory constrained than Original with lowmem. Generally you'll get fastest results with a higher BS. While BS=1 may go through iterations faster, it's actually slower than a higher BS which can go through ...
by bryanlyon
Tue Mar 16, 2021 7:39 pm
Forum: Training Support
Topic: Doesn't seem to use gpu
Replies: 2
Views: 880

Re: Doesn't seem to use gpu

In addition to what torzdf said, I'd like to point out that usint a 2gb card is going to be very hard. At best you'll be able to pull off one of the smaller models. You may want to see if you can get a larger card or use cloud services.

by bryanlyon
Tue Mar 16, 2021 7:34 pm
Forum: Training Discussion
Topic: RealFace or Dlight when training for 128px output?
Replies: 7
Views: 1706

Re: RealFace or Dlight when training for 128px output?

It's more complicated than that. The encoder is trained to take a face and encode it so that the decoder can re-create the face. Giving it a 64px or 128px it should be able to get all the face data it needs to be able to feed the decoder. The decoder is responsible for all the detail above that.