This is a PhotoBooth style transfer workflow. You begin by inputting a photo, such as a webcam photo taken in portrait mode in the booth. Next, you input the style transfer photo, which could be a specific outfit and style (like a photo from the Joker movie or the Mona Lisa), and finally, a background (which could be the empty booth or a specific location). We retain only the primary subject, as the booth may capture other people in the vicinity.

To change the clothing with the style input, we use a silhouette mask of the outfit instead of the Canny of the main clothing. We use Cloth Segmentation techniques to do that. This approach helps avoid any specific patterns or multilayers from the clothing.

We employ OpenPose and Canny for the details of the face, hair, and posture, aiming to produce a result that closely resembles the input photo’s face.

To enhance the result, we conclude with a DeepFake of the original face on the resulting image.

Enjoy the experience!