Faceswap Forum

Blurry output can happen for a lot of reasons. Not enough variety in the data (this is the most important one. Data should be highly varied) Resolution of the model (IAE is 64px. It is always going to be blurry to an extent) Model needs to have warp disabled (see here: https://forum.faceswap.dev/vie...

Hmmm. Not seen this before, and really don't know what the issue could be. RandomNormal is a standard initializer which definitely exists in Keras: https://github.com/keras-team/keras/blob/d8fcb9d4d4dad45080ecfdd575483653028f8eda/keras/initializers/initializers_v2.py#L302 A quick bit of googling sug...

No. This should happen automatically. The process reads the input's frame rate. If it is outputting at an incorrect framerate, then most likely it is misreporting. I would suggest using "open-cv" writer to output converted frames, and then use ffmpeg (either the Effmpeg tool in Faceswap, o...

I have pushed an update to revert this change. I will probably need to revisit in the future, but it should solve your issues for now.

Unfortunately this is as about as far as I can help. The fact that you are installing cudatoolkit and it isn't showing up suggests something weird with the setup on the cloud image, which is outside of anything I can help with. The fact that it does find cuda but reports linking to the wrong version...

Without knowing the exact command you are running, the quality/size of the source video or how many faces are in each frame it's impossible to know. All of these things will impact extraction speed in some way. The biggest things you have control of is how many plugins you are running at the same ti...

How can you tell from the sysinfo that they are not there? The top level system info for my old environment (which still works for training, just not for extract) shows basically the same info. I don't see any install errors. It unpacks cudatoolkit and cudnn and happily installs them. They are not ...

Note there's no global cuda found, which I think is the thing that most people get tripped up by. Also, in the sysinfo, it has no problem seeing that there IS a GPU, so... help? Yes, that's what you want. However, there should be a cudatoolkit and cudnn installed locally in your conda environment, ...

You're basically running out of system memory. Try reducing the number of jobs running (near the bottom of the convert tab)

You have an Nvidia GPU but you have installed Faceswap for AMD.

Delete the Faceswap folder and reinstall for Nvidia

Please post the full crash report from the faceswap folder when posting issues.

However, I'm pretty sure here that you have images in the timelapse folder that does not exist in your training folder.

The main problem with this is that the numbering is entirely arbitrary, and interrupting the convert process for user input not straightforward

I don't know if it will be better, it may be or it may not. I've never really tested at length so don't know how much the encoder information gets rewritten. I certainly no of people who do what you are saying though and just keep re-using their encoder.

I can confirm this effect of swapping only one face per frame/picture if multiple faces are in the alignment file. I also found the root for this effect: lib/convert.py in line 248 the following change was successful: "borderMode=cv2.BORDER_CONSTANT" -> "borderMode=cv2.BORDER_TRANSPA...

On, convert, yes. Full frames take a lot of RAM, You can try reducing the number of parallel processes to use less RAM.

No it is not. A model is defined based on its input/output size. You cannot change this for saved models. You will need to create a new model with a larger input size and start again.

you need to install git.

The easiest way to do this is to run the windows installer as it does this for you:
https://github.com/deepfakes/faceswap/r ... tag/v2.0.0

app.php/rules#rule-4a

Is this the correct situation to use load weights + freeze weights? It is a correct situation, yes. Other people re-use their trained encoder to kick start training totally new identities too (say, C + D). Effectively the logic is "you can use weights from a different problem to solve this pro...

Yes, this was a bug. I have pushed an update which should fix this.

Faceswap Forum

Search found 1760 matches

Re: output face is blurry

Re: Training Continuation Error: Unknown config_item: RandomNormal

Re: mismatch frame rate (output & original)

Re: Not converting multiple faces in one frame

Re: GPU not being recognized though it's definitely there (even visible in sysinfo)

Re: Why is extraction slow

Re: GPU not being recognized though it's definitely there (even visible in sysinfo)

Re: GPU not being recognized though it's definitely there (even visible in sysinfo)

Re: numpy.core._exceptions._ArrayMemoryError: Unable to allocate 31.6 MiB for an array with shape (1080, 1920, 4) and da

Re: Unsupported Operand Type(s) Error

Re: Critical error

Re: Not converting multiple faces in one frame

Re: Is "load weights" the right way to go if I only need the B part?

Re: Not converting multiple faces in one frame

Re: Critical Error on Large Files

Re: [DFL-SAE] Is it possible to change the input resolution during training?

Re: Amd GPU cannot open GUI

Re: Unsupported Operand Type(s) Error

Re: Is "load weights" the right way to go if I only need the B part?

Re: Unsupported Operand Type(s) Error