[Feature Request]: FaceSwap optimizer

[kalle07]

Is there an existing issue for this?

• I have searched the existing issues and checked the recent builds/commits

What would your feature do?

all is working but ...
the weight of the whole image is taken too much
at the moment only PNG with deleted areas are usefull to erase all except the face

Proposed workflow

...

Additional information

so my wish is ...
for option FaceSwap in inpaint ... run first a face-detection and than only this region is taken for the face region in the generated image, so further its useful to doo a face-detection with low-level in after ~20% of steps (in the image that is generated) to overlap these two areas more exact ...

[mashb1t]

So sorry, but I don't get it. Can you maybe rephrase your request? ^^

[kalle07]

hmmm...
ok, i write a prompt "portrait of a woman wearing a blackshirt"
and i take an image of a woman and mark it for faseswap (this woman wears a white shirt) ...
so if i generate the image mostly a white shirt apear ...
so i dont know how faceswaper works but seems the whole image is taken not only the face-region.
and my idea in adition was to recognize as fast as possible the face in the image that is generated.
if i take 60steps i think at step 20 with 30% precision (eg. yolo) can recognize the location of the face, so also there can be calculated more exactly...
it is now a little easier to understand?
"at the moment only png with erased regions (not black or white painted, erased, so only the face is visible) are working with a color of the top i write in the prompt", i can send you such an image if you like