Faceswap Forum

hyln wrote: ↑Fri Mar 22, 2024 11:50 am

Once I start a training, I got below error from my crash report, how can I fix it

You input images on the B side has non faceswap extracted images in it. The training/timelapse images must all be faceswap extracted faces, and should be the same size.

Unfortunately there is not enough information in that error message to know what the issue is.

Most likely you are missing one of the pre-requisites

There was a bug which prevented the installer from running.

I have just pushed a fix for this, so please try running the installer again (no need to re-download it).

Most likely you will need to delete your model, lower the learning rate (try 3.5e-5, if that doesn't work, then 1e-5) and start again.

Most likely one or both of:

• D:/faceswap/A/faceimagesA
• D:/faceswap/B/faceimagesB

does not contain faceswap extracted facesets. All images in these folders must be face images that have been extracted by faceswap

What do you mean nothing works?

Try:

conda activate faceswap
cd faceswap
python faceswap.py gui

Post any error output

I can honestly say I've never seen this issue. I've had models where eye-tracking has gone wonky, but not your specific example.

Sorry it took so long to get around to this.

Unfortunately I was unable to recreate the issue on either a Windows machine or a Linux machine, so it is most likely a problem specific to AMD. As I do not have an AMD card, unfortunately I cannot investigate further.

This should be fine, assuming your data quality is good.

You don't train a faceswap model to swap a scene. You train a model to swap 2 identities.

Once you have a trained model on 2 identities, you can use that model to swap the faces in any scene where the A face appears.

This is a system issue, in some way (a required file missing on your PC, or a conflict of some kind).

In the first instance, try this to see if it resolves it:
app.php/faqpage#f1r1

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.