Make sure the two changes I made a few minutes ago actually work (waiting on a per-megascan ZIP file extraction over a few hundred ZIP files at the moment).
Figure out a better way to hunt down the megascan python file path. Or just add a command to the shell in the powershell profile.
The general format of each 'recipe' is something like: "Take a bunch of files and bundle each one into a varying number of channels."
Examples we have:
_n.tga is produced from *Normal.jpg (mandatory) (one to many)
_a_d.tga is produced from *Albedo.jpg mapped over RGB and *Displacement.jpg mapped over A (one to many + one to one)
_m_r_c_ao.tga is produced from *Metallic.jpg (optional), *Roughness.jpg, *Cavity.jpg, *AO.jpg mapped to R, G, B, A respectively (one to one)
We can generically implement one-to-one and one-to-many recipes and build a parser that reads an input datafile for any arbitrary demands on this system. That way, we can support new bundle formats as they are required (I think there is at least one new one that Trent needs that this one does not currently support, though it is quite rare).
Turns out the approach to the FFT kernel generator was woefully... Wrong. I have to figure out how to even approach properly generating one of these things.
The megascan assembler uses "family root" for the root of the output filename, and "filename root" for the root of the input filename. These are not great names, they should get changed.