Hmm, I can’t really agree on using IOR as input being better. Reasons:
- The effect of going from 1.45 to 1.55 where most dielectrics reside, is absolutely minimal wrt difference in reflection output. Sure, glass and diamonds are up there, but what sells the effect? If you made diamond with IOR 2.0, do you think anyone would notice or care? Bending light a lot and being very brilliant, in combination with a decent environment is what makes the image better (ignoring what we can’t have wrt rendering diamonds).
- Dealing with values > 1 are more cumbersome to preview, as at any step in the process of manipulating it you have to add additional math nodes to make sense of the visual preview (using the emission node).
- Many times you will find specular textures in a range that already makes sense, or at least a much simpler 1-n math operator is the only thing needed (metallic/rough hasn’t been around forever, nor do their textures).
- With Disneys goal in mind, think like an artist, not like a scientist. Strive for what looks good rather than going for realism for realisms sake.
- With above in mind, imagine floor boards with tiny gaps in between. Light would “get lost” in these cracks and eventually get absorbed, producing almost no visible diffuse or specular output. Hooking up the “grout/mortar map” (?), to me it feels easier to preview and scale the b-w image to say 0.05-0.5 rather than having to convert it to IOR, then try to manipulate it with maths dealing with >1 values.
- For glass (and topcoats; roughness bug still in 2.80a Cycles, but fixed for Eevee), I don’t really use Principled anyway. Vanilla glass shader is also still very lacking.