This is semi-custom implementation of faceswap as a Cog model, based on the original code of roop. Cog packages machine learning models as standard containers.
First, download the pre-trained weights:
cog build -t faceswap
Then, you can run predictions:
cog predict -i target_image=@tony.jpg -i swap_image=@elon.jpg
Example output for sample inputs:
