This is a python script which removes the backgrounds from images, centers the image, and then resizes the image to a square. I made this to automate Juliet Siegel's workflow when making filters in EffectHouse.

./filter.py <desired-resolution> <input-folder> <output-folder>

When formatting ~700 images during the creation of one of Juliet's outfit filters, this tool succesfully converted about 600/700 with an acceptable quality. Roughly an 85% success rate. It struggled on images where:

• the subject contained similar colors to the background
• the subject is very thin or stringy (for example, if the subject of the image is a necklace or wristband).

No ML tool is perfect, and this script was cobbled together in a few hours using off-the-shelf libraries. Improvements could be made by improving the rembg python library.

See the source code lol