It's open source ;) Apparently, they extract video frames using FFMPEG, create a collage out of those frames, then use the whash method of the python imagehash

It's open source ;) Apparently, they extract video frames using FFMPEG, create a collage out of those frames, then use the whash method of the python imagehash package. So it's basically reducing video hashing to image hashing, which was previously solved.