Your file, as video frames
The v5 pixel codec, running on what you give it. Two bits per 4×4 block of luma and three per block of chroma, 56,100 bytes to a 1080p frame. Encode it, decode it, compare the two.
What happened to your bytes
Eight steps, four down and four back. Every number and every block value here came out of the round trip you just ran.
The frames
One image pixel per 4×4 block. The colour is not a tint: all three planes carry payload, so a grey thumbnail would be showing you a bit under 58% of the data and hiding the rest. At full size these are 1920×1080, and what makes them survive a transcode is the distance between levels, not the detail.
What is not here
This runs the middle of the pipeline. The full project archives and encrypts first, protects the result with Reed-Solomon, encodes the frames to H.264 and uploads them to YouTube unlisted — with the header written three separate ways so a decoder can find it even if the transcode eats two of them.
None of that belongs on a public URL: uploading needs credentials for one particular account, and H.264 plus a browser session is more than this server has. The repository runs the whole thing locally, and there is a Docker image and a Windows executable for doing it without a Python install.