Faceswap Forum

Right now, if no masks work for you there isn't much you can do about it. Torzdf does have an updated manual tool in the works that will allow you to manually mask the face, but right now it is not yet available.

This is often masking a memory error. If you've already turned on Allow Growth in the Extract plugins, you may try turning down the BatchSize in the Settings/Extract Plugins/Mask/Unet DFL.

viewtopic.php?f=6&t=146
This is the best source of information on this. Honestly, much more than what this guide shows is overthinking things. Just go for the largest Batch Size you can without turning on any memory hacks (as those slow down training for Vram).

Training speed is not just iterations. It's actually best to use the number of images in a time period. So Batch Size * Iterations is a better calculation So 50 * 240 = 12,000 images/hour 30 * 500 = 15,000 images/hour 6 * 900 = 5,400 images/hour (I however, believe you've likely mistaken the speeds ...

Not really, no. Storage is much slower than the vtam on board the gpu and relying on it as vram replacement is not viable. Also, the images must be augmented so even streaming the raw images isn't viable. Further, consumer pcs don't have separate pci-e switches that allow the gpu to directly read fr...

This is a weird error that we've seen once before. See https://github.com/deepfakes/faceswap/issues/954 Unfortunately, none of the developers have been able to reproduce it. We are trying to look into it, but it may simply be a corrupted install. That error code is not coming from our application, b...

Please post your crash log (if present). It'll be in the faceswap folder named "crash_report[date].[longnumber].log . If not, post your system info: GUI Users: Go to Help -> Output system information CLI Users: From inside your virtual environment, inside your faceswap folder, run: python -c "from l...

Any time there is a mark you want to preserve on one side of the face you should disable the flip option throughout the entire training process. If you do this please be aware that you'll want to include a significantly larger dataset to ensure the quality of the results.

It's not critical. You could certainly get away without upgrading the CPU, but you *WILL* see better speeds if you upgrade it later. Might be worth scheduling it sometime down the line.

Yes. a 2060 super would let you turn off a lot of the memory preservation you're currently using which is honestly the thing most slowing you down with your 960. I'm not going to promise double the speed, but you'll see some strong speed improvements. The only problem is that with a 2060 super, your...

I've never used it. They *look* fairly legit but you can never be sure. Maybe you'd rather try the free tier of Paperspace https://docs.paperspace.com/gradient/in ... -instances ?

From the faceswap folder, plugins/extract/recognition/.cache/vggface2_resnet50_v2.h5

Villain is barely able to run on 8gb of vram. Try a BS of 1. If that doesn't work, try closing extra programs (Like browsers). Unfortunately MSG and other memory saving features come at a cost of time, that's just the tradeoff that you have to make to run larger models with limited memory.

No, leave it at 256.

It sounds like you were marginally capable at that BS. Sometimes just by luck you run at a batch size once that wont work again later. Try a lower BS or close some applications that may be using vram (Browsers and Video Players are good examples).

Flash, probably, but you should definitely use set lights. Lighting is probably the most important thing that people miss out on.

It's important to remember that there is a difference between Training data videos and the swap video. Making a quality training set needs a huge variety, but the swap video you want to minimize complexity in any way you can. For training data, just run the camera, catch video when not actually film...

You want the training faces to be at least that 128x or you're training to lower quality faces and you wont get full quality for the swap.

Villain is already a very heavy model. Without some exotic hardware with a lot of vram, it's pretty much as big as it can get.

1. "with 100% coverage and extended mask" See https://forum.faceswap.dev/viewtopic.php?f=6&t=146#config 2. "pretrained the villain model for 50k iterations on random faces" why would he do this...or did he mean random faces of Keaunu ? There is some debate as to whether pretrain is useful or not. P...