Hello,
We are considering to use MDL for rendering AXF materials and I have noticed that the import of carpaint2 AXF representation is not very accurate - namely the flake size and density is much lower. If I understand correctly, it is because in the AXF format the flakes are represented as a BTF but in MDL they are procedurally generated, so the original BTF is not used in any way.
Do you have plans to improve the import of carpaint so that the flakes match the original BTF?
Can you suggest a way we could work around this? Maybe by creating a custom importer and MDL material with texture_3d to represent the BTF?
Are you aware of any other properties of materials that are similarly not very well preserved during import? (especially in SVBRDF and Carpaint2 AXF representations)
The procedural flake model is a very accurate fit of the flake BTF (we numerically fit the density to the density in the btf), which imho improves on some of the shortcomings of the BTF (limited resolution, repetition artifacts, potential mismatch of near and far-field material look, unrealistic smoothing/filtering of flakes).
Also, it would be important to note that the renderer-used also plays a role in the perceived “density” and intensity of the flakes. For high specular flakes, many renderer will clamp intensities to reduce noise. This will happen independently of whether BTF or procedural flakes are used. We added artistic controls to the axf_carpaint material to allow users to compensate (flake_roughness, slight increase will probably be perceived as higher flake density if lit with high intensity lightsources)
We have currently no plans to support BTF and we dont see BTF as an improvement here, especially in the context of accurate light simulation.
A BTF is more then a 3d texture, you could potentially derive a normalmap, an intensity map and maybe a roughness map to match the “filtered” look of the btf flakes, we have not explored that path ourselves.
The other known limitation is, if you use “refractive clearcoat”, we discourage exporting your axf materials with the “refractive clearcoat” option.
Thank you for the information. I agree that the procedural model is superior.
I have experimented with multiple renderers, varying flake roughness, TMOs, disabled clamping / firefly filter while using the same environment map with a small high intensity hotspot, however there was still a noticeable difference in flake density, glint spread and glint color compared to the original AXF material.
So for the moment it looks like we will have to implement the AXF shader as well.