Faceswap Forum

There was a bug introduced by a Miniconda update. This should now be fixed, which should resolve the most recent install issues

No, downgrading Python would not work.

The Dockerfiles have never been tested for macOS, so they may not work (and haven't been tested more widely for a long time anyway)

Try a clean install (app.php/faqpage#f1r1)

There was a change in Miniconda recently that broke the install script I pushed a change yesterday that will hopefully have fixed this

It is not easy, no, as any edits you make to the extracted faces will strip the metadata from the header that is required by faceswap.

See here for a user submitted guide for how you may be able to resolve this issue:
viewtopic.php?t=1733

Both have their benefits, and a high quality dataset is likely to be sourced from both types of images.

The downside of images is they tend to be posed, so whilst quality is normally greater, variety in pose, expression (especially blinking!) and lighting tends to be far worse.

Try doing a clean re-install (see here: app.php/faqpage#f1r1)

Conda made a change which broke the install script, and I pushed an update to fix this yesterday. Hopefully that will solve your issue

No. VRAM is VRAM. You can either fit a model into it or you can't.

It sounds like you are looking in the wrong place.

The model saves:

• Every 250 iterations (by default) or whatever value you set
• When you stop training.

This looks like a bug. I have tagged it as such and will investigate when I have time. If you could provide a preset of a model which fails in this way, that would be useful.

We do, but I have not had time to work on this.

Keep an eye on this thread as it is where the update will happen:
viewtopic.php?t=3010

This was a regression caused by an updated in how MiniConda handles requirements.

If you re-install faceswap, it should now work.

If you still have issues please provide the newest faceswap_setup.log from your faceswap folder

app.php/rules#rule-4b

In your example:

A should be the input video. The alignments file for A is the one you should use. No need to provide a reference video, as A is already a video.

You are most likely pointing at an incorrect model folder.

Faceswap creates a model state file when training. The folder you are pointing at has none, so the convert process cannot continue.

Looks like your input is pointing at a faces folder?

This looks like a problem with imageio-ffmpeg. We have downgraded the version we use. If you reinstall faceswap, hopefully the issue will be fixed.

You are going to need to be more specific.

Please detail the exact steps you took to get to your issue.