Authoring reliable metallic-roughness PBR materials

Treat inputs as material information

A metallic-roughness workflow is useful because a small set of inputs can describe many opaque surfaces. It still needs disciplined authoring. If a texture contains highlights from the photograph used to create it, the runtime light adds another highlight. Begin with a material hypothesis and a controlled inspection scene, rather than adjusting maps until one screenshot looks attractive.

Understand the parameter contract

In the standard model, base colour describes dielectric diffuse colour or metallic specular colour. Metallic selects between these responses; roughness changes the microsurface reflection distribution. An energy-aware shader coordinates diffuse and specular terms so reflection does not create energy. [1]

File conventions matter. glTF stores base-colour RGB with the sRGB transfer function, while its metallic-roughness texture contains linear data: roughness in G and metallic in B. [2] Other pipelines can use different packing; record the actual import contract.

Work through a painted crate

For an illustrative steel crate, separate blue paint, exposed scratches, and rust. Start with metallic 0 for opaque paint and rust, and metallic 1 for exposed steel. Try roughness 0.65, 0.25, and 0.9 respectively. These are authored trial values, not measured material constants.

Inspect the scratch mask close up and at gameplay distance. Intermediate metallic values can represent filtered boundaries between regions; making the entire painted panel metallic 0.5 instead changes its response everywhere. Keep a copy with constant roughness to distinguish a classification error from excessively busy surface detail.

Establish a repeatable procedure

First inspect every imported channel as greyscale data and compare it with the source image. Then rotate a broad light around the asset with exposure fixed. Check the same material under an environment containing both bright and dark regions; a uniform environment conceals reflection structure.

Disable normal detail temporarily, then restore it after the base response is stable. Compare a smooth reference sphere beside the asset. Finally export and reimport a small test material, verifying channel packing and colour interpretation before processing the full asset collection.

Know what the model omits

A compact material model cannot reproduce every measured surface. Disney's original development explicitly balanced observed material behaviour with manageable artistic controls. [3] Layered coatings, transmission, anisotropy, and subsurface effects may require additional models.

Do not compensate for missing illumination by painting it into base colour. Record which appearance is a material property, which comes from lighting, and which is an intentional artistic approximation. That record makes future renderer changes easier to diagnose.

Sources

• [1] Physically Based Rendering in Filament — Google Filament.
• [2] glTF 2.0 material specification — Khronos Group.
• [3] Physically Based Shading at Disney — Brent Burley, Walt Disney Animation Studios.