search

cubiq / ComfyUI_IPAdapter_plus

GNU General Public License v3.0
3.76k stars 286 forks source link

Keep face consistency with multiple IPAdapter Face using attention mask #285

Closed julien-blanchon closed 5 months ago

julien-blanchon commented 6 months ago

I'm trying to experiment with FaceID to keep the face consistency when doing Image2Image.

Basically I'm transforming this input image:

Screenshot 2024-03-03 at 00 01 56

In this:

Pixar3_00195_

With an Image2Image fashion.

I'm trying to use IPAdapter FaceID to keep the face of the two person, and possibly to any given number of person in a photo.

When using only 1 IPAdapter this either make a merge of the two face or put the same face on the two persons. I would like to be able to use the SEGS/BBOX from a face detector to generate the needed number of IPAdapter with the attention mask given by the SEGS/BBOX itself. However when I give batch/list of image and attention mask to IPAdapter this consider it as multiple instance of the same face and failed to give good result. When I enable unfold batch I get multiple model.

cubiq commented 5 months ago

closing for inactivity, unfortunately I don't have time to also offer workflow support