Any plans to fork FaceFusion instead since roop is dead?

[UberGuidoZ]

Since the main dev of roop has decided to close the project and no longer develop it, are there plans to change the main fork over to the successor, FaceFusion?

[C0untFloyd]

No. The last month this fork had so many advancements over the original roop and face fusion, I don't even think of it as a fork anymore. If there is something face fusion is doing better or a bug fix which affects both, I could still do some cherry-picking. I would rather fork from Rope RIP as I feel it had the better features and more advanced stuff.

[alast0r]

this is so true this is right now the leading engine for faceswap the others are not even close great work mate

[tfis2]

Rope seem to be alive
https://github.com/Hillobar/Rope/tree/main

[C0untFloyd]

Wow, back from the dead! Great, I always liked Rope but disliked the GUI/Usability.
Funny that everybody (even Rope) seems to "borrow" from everybody else:
Merged Source Faces._
Select multiple Source Faces to create an average Source Face result. Combine different people to create a blend, or select the same person from different Source Faces to improve the likeness of the person

This is something one of the SD Extension did first, then Swap Mukhamsomething stole it and now Rope 😄
I need to steal the blending code from Refacer (which is deactivated now) and then I can rip that thing off too (would be needed for DMDNet specific anyway). Perhaps I can be the one who does the best masking first...

[aripsam]

Yes, I think you’re referring to faceswaplab. You can select multiple faces to create a blended model.

Has anyone ever tried this -

Source image is a side view of a face, target is also side view of a face. Will the side face source work better than a front view face for a side view target?

If yes, maybe there can be logic that detects face alignment and you can supply around 4-6 source images at various alignments, and that improves the results?

[C0untFloyd]

Has anyone ever tried this -
Source image is a side view of a face, target is also side view of a face. Will the side face source work better than a front view face for a side view target?

Doubting this. All it does is averaging the facial positions which would stabilize it but it's unlikely it would improve side views or extreme angles. You can blend face features of multiple persons together though, I personally see no use for this though.

[UberGuidoZ]

Awesome, as long as development continues I'm a happy camper. I did see FaceFusion is adding a live swap on webcam which is pretty nifty. Looks like Rope has that as well. Very excited to see where this goes!

[sordidloam]

Better yet.  Choose a front photo and a side photo and the swapper create a truly accurate depth of the face.Sent from my iPhoneOn Aug 31, 2023, at 6:56 PM, aripsam ***@***.***> wrote: Yes, I think you’re referring to faceswaplab. You can select multiple faces to create a blended model. Has anyone ever tried this - Source image is a side view of a face, target is also side view of a face. Will the side face source work better than a front view face for a side view target? If yes, maybe there can be logic that detects face alignment and you can supply around 4-6 source images at various alignments, and that improves the results? —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[huyzer]

"Better yet. Choose a front photo and a side photo and the swapper create a truly accurate depth of the face"

Is this something you've tried and compared side-by-side, or a hunch?

[ssuukk]

I used roop by combining multiple photos of the same person's face in different orientations, it was unfortunately much better than anything I could obtain from roop-unleashed.