Honestly, it really shouldn't be that hard, just grab the bitmaps, encode them, and send them over a socket. I suspect that the approach that NTR CFW is taking isn't actually ideal, it just JPEG encodes the bitmaps individually and sends them across. This is maybe quicker than actually encoding video, and it definitely makes the code simpler, but you might be able to do better using an actual video codec like AV1 or even just using a better image codec like JPEG XL. The problem is that you only have a few cores (probably 2 at most) and no NEON instructions, so everything has to be done in software. It might be interesting to just port the approach used in NTR, but make it more modular so that different encoding techniques could be experimented with (which would also be a lot simpler if it was a plugin instead of a bulit-in part of Luma/Rosalina). Thinking about it, there are a number of parts of Luma that could potentially be broken out into plugins already like the Cheat Engine and input redirection.