nyn-05/major_project ? reverse-engineered prompt

Build me a deepfake video detection tool for KYC videos.

I want it to take a short video, sample the frames, check face quality, extract the rPPG heartbeat style signals from the face, and then use that data to decide if the video looks REAL, FAKE, or UNCERTAIN. Please wire the whole flow together so I can run one command and get a result JSON plus any useful outputs saved along the way.

Also include a simple web interface where I can upload a video and see the verdict clearly, along with a basic explanation of the signal quality or why the clip was rejected if it was too blurry, too dark, no face, or too small. Keep it practical and lightweight, and make sure the pipeline works with the sample video in the repo. If you need to, look up current docs online for any libraries that are required.