I understand that the goal is to make the data survive video compression, but wouldn't it make sense to use at least some color information instead of entirely black and white pixels?
there should be some error correction in a system like this, though.
QR codes have 4 levels of correction you can use depending on how robust you wish them to be. CDs and DVDs use two chained, fixed, levels to keep the decoders simple. CDs have 25% overhead, but their correction is very strong: they can correct 4000 bits in a row.
There are two encoding modes, RGB and B/W. It uses a pixel-to-data width of 2x2, but says YouTube's compression algorithm is brutal, and one corrupted pixel already renders the whole thing corrupted.
Now, if you're using a 4:4:4 format to do this, then you should be able to use smaller chroma metapixels (I still wouldn't use the full chroma resolution, though, unless you're using a high bitrate or a lossless codec). However, that risks data corruption if passed through a pipeline that downsamples the chroma.