Workflow for swapping face of one person by creating 1024x1024 image with similar style and pose (IPAdapter) and face (InstantID) than overlaying both images. Includes SegmUltra for automatic mask generation, but often better results can be achieved by manual mask on top of automatic.
Tips:
try changing prompt in SegmUltra ("head", "face")
if face is small in original image try increasing megapixels in resize (default is 1 for portrait)
muting groups works to tweak masks, load image manually into load image nodes
V2 has major changes to workflow:
Automatic prompt from Florence
Expression editor for manually correcting face expression
Tweaks for many settings
Description
Major changes to workflow:
Automatic prompt from Florence
Expression editor for manually correcting face expression
Tweaks for many settings