You can extract each frame from the input video and identify white background pixels using color thresholding in the HSV color space. and then create an alpha channel is by marking the white regions as fully transparent and the rest as opaque. Save these processed frames are PNG images with RGBA channels. Finally, encode the sequence of transparent PNG frames into a .webm video using the VP9 codec with specific FFmpeg settings that preserve transparency, resulting in a video where the original white background has been removed and replaced with true transparency. You can refer to: https://colab.research.google.com/drive/1Jz39wRN4hiJbvGsrOTFYP4uymjhOlJgF?usp=sharing