BRDFs as style choices - why it matters for art direction

‘I can barely tell the difference’ was the response I got when demo’ing the change between Oren-Nayar over a standard lambert diffuse. While true, the difference can be seen as minor, it makes a bigger overall impact and also adheres to roughness/gloss.

Oren-Nayar going from 0 to 0.3 roughness. At 0 roughness, we essentially achieve the same look as a lambert.

So for the lead artist or art director who is not familiar with a BRDF (bidirectional reflectance distribution function)- what is it, and why should you care?

In layman terms, a BRDF is a function that calculates how lighting is distributed on a surface. Some are developed to replicate certain surface, like Minnaert that is commonly used to mimic cloth-like surfaces such as velvet (although it was originally intended for the moon’s surface).

For starters, the shading model of a game can have drastic effects on how both the end look is achieved, and how the assets that need to be created look like. Below we can see a screenshot of The Legend of Korra by Platinum Games. It blends banded characters on hand painted backgrounds. This cel banded look tends to the most noticeable, and most common deviation from traditional diffuse shading models:

A quick look at part of the texture on the character above, and we can see how simplistic the color information is. Here, the reliance is on both the lighting the mesh geometry to create the extra detail. There is, of course, no particular reason why you could not have the lighting and banding be affected by both the values of the texture map and the contour of the geometry.

You can mix and match diffuse and specular models (and all games do) with decent result and differences. In Guardians of the Galaxy: a Telltale Series (below), we can see multiple BRDFs in action, working harmoniously to achieve a believable yet stylzed end result - from the hair (Kajiya-Kay), head (Oren-Nayar/GGX) to the sweater (RAD cloth).

Below, we can see a much more advanced diffuse model providing cartoon like shading.

Even though it is 'cel-shaded' like Korra, the addition of specular, and normals contribute to breaking up the typical cel-banding - giving the final look a more organic feel.

Or we can rely on just the art assets to provide the style, and stick a more common set of BRDFs. Life is Strange: Before the Storm and Walking Dead Season 3 do this to good results, and the final image and styles are distinct as the art direction ends up separating the two. Below we can see two characters, with drastically different looks thanks to textural, lighting and modeling differences. 

As it stands, and from the examples shown above - I hope it is somewhat clearer to art directors and leads reading this, that the shading model that your project is using is important in achieving the end result of a look, even while stylized. There are many established BRDFs out there, and starting a dialogue with your engineering and tech art team in pre-production can help get better results, or different results. I mentioned it briefly in a previous blog post, but one of the tools I found to have good success with was Disney’s BRDF viewer - a simple application that people can use to evaluate how light reacts to a surface with handy sliders to control attributes. While not perfect in terms of usability, it does provide a decent option to compare and evaluate what might work for a project.

Something that I’ve struggled to show art directors and artists in the past was showing what the perceived changes on a test asset would look like, and side by side comparisons. Most artists are not aware of the differences or simply do not care - but they should as the smallest change on an overall character or scene can shift the look quite a bit. Perhaps look development, a practice often associated with films and animation, should be considered more heavily by games.

One thing to keep in mind - there is a solid ground truth that most real-time shading models will attempt to achieve. However, due to runtime performance, approximations are made and performance vs accuracy trade-offs are decided upon. To assume that one viewer, say Marmoset or Sketchfab, is the ‘right’ one is naive and the final result should always, always be looked at in your studio’s or project’s engine of choice. Content made for the game does not live in a viewer for an artist’s portfolio but rather in the game itself.

Special Thanks for Matt Davidson, who helped introduce and show me the importance of BRDFs. The look and feel of many of Telltale's newer projects would not have been possible without his work.