Ok. It seems there was a requirement update on PlaidML which broke our dependencies.
I've pushed a fix, so hopefully this will fix it for you. Please clean everything out and try installing again.
Ok. It seems there was a requirement update on PlaidML which broke our dependencies.
I've pushed a fix, so hopefully this will fix it for you. Please clean everything out and try installing again.
Ok. Basically PlaidML is failing to load an internal file....
The next thing I would try is the following....
Start > Anaconda Prompt
Code: Select all
conda activate faceswap
plaidml-setup
Follow the prompts to select a GPU, then try Faceswap again.
The most likely cause is a conflict somewhere.
I'm hoping our AMD guy shows up soon with some troubleshooting, but could you try to follow these steps and see if it resolves the issue?
Ignore it. They are internal Keras warnings.
Whilst this isn't currently directly an option in Faceswap, you can do this by setting an environment variable.
See here:
https://stackoverflow.com/questions/497 ... indows-cmd
Should be fine. You can also add additional masks with the mask tool
Can you post the raw output from this command:
Code: Select all
C:\ProgramData\Anaconda3\envs\faceswap\Library\bin\ffmpeg.exe -i C:\Healthy\faceswap\input\input_vid1.mp4 -map 0:v:0 -c copy -f null -
Ahh, that may be reserved RAM, rather than explicitly in use. Are you looking at the free column or the available column?
The total ram less available will give you a better indication of the amount of RAM actually being used.
For a first fake, that's very good. My first was far worse! You made a decent choice for source video for a first swap (mostly facing forwards)
14GB of system memory? That doesn't sound right to me.
I have honestly never seen this message before, and Google isn't turning much up. I would suspect that it's an out of RAM/VRAM issue, but I may be way off.
There are tips for knowing when the model is ready here: viewtopic.php?f=6&t=146#monitor
Villain will run on an 8GB card, but it will be tough (maybe impossible?) on Windows because of the OS overhead.
If you can, then I run Linux (Xubuntu would be my recommendation), maybe via dual boot, to get access to more of your card's VRAM.
You wouldn't really want to use a combined alignments file for that.
You would generate an alignments file specific to the video you're wanting to convert.
Unfortunately the information we require has been truncated from the Crash Report.
If possible, post it on pastebin (or something similar) or add it as an attachment to the message
Just as an update to this.
All masks are now available in convert, and I don't recommend learning a mask. Just make select the required mask at convert stage.
You have passed in your model's state file rather than the frame's alignments file.
We would need to crash report.