Loaded up the Digital Humans sample to look at it, expecting to see the same problem… well…
It’s using the subsurface profile shading model, the subsurface scattering setup is painfully simple, they’re just combining masks with lerps, using params for the lerp alpha so that they can control the scattering for all the individual areas of the face.
Clearly I am missing something but I don’t know what
Edit: All the values in the SSS are set above 0.1 and there’s practically no detail in them, they’re just big soft blobs. I’m guessing the shading model does some kind of volume approximation for the backscattering so it’s not necessary to mask out areas where you would expect no backscattering to occur.