The mental ray mia_material is a monolithic material shader that is designed to support most materials used by architectural and product design renderings. It supports most hard-surface materials such as metal, wood and glass. It is especially tuned for fast glossy reflections and refractions (replacing the DGS material) and high-quality glass (replacing the dielectric material).

The major features are:

The material comes in two variants, the mental ray 3.5 compatible mia_material and the new extended mia_material_x. These are just two different interfaces using the same underlying code, so the functionality is identical, except that mia_material_x...

The mia_material primarily attempts to be physically accurate hence it has an output with a high dynamic range. How visually pleasing the material looks depends on how the mapping of colors inside the renderer to colors displayed on the screen is done.

When working with the mia_material it is highly encouraged to make sure one is operating through a tone mapper/exposure control or at the very least are using gamma correction.

Describing all the details about gamma correction is beyond the scope of this document and this is just a brief overview.

The color space of a normal off-the-shelf computer screen is not linear. The color with RGB value 200 200 200 is not twice as bright as a color with RGB value 100 100 100 as one would expect.

This is not a "bug" because due to the fact that our eyes see light in a non linear way, the former color is actually perceived to be about twice as bright as the latter. This makes the color space of a normal computer screen roughly perceptually uniform. This is a good thing, and is actually the main reason 24 bit color (with only 8 bits - 256 discrete levels - for each of the red, green and blue components) looks as good as it does to our eyes.

The problem is that physically correct computer graphics operate in a true linear color space where a value represents actual light energy. If one simply maps the range of colors output to the renderer naively to the 0-255 range of each RGB color component it is incorrect.

The solution is to introduce a mapping of some sort. One of these methods is called gamma correction.

Another method to map the physical energies inside the renderer to visually pleasing pixel values is known as tone mapping. This can be done either by rendering to a floating point file format and using external software, or use some plugin to the renderer to do it on-the-fly.

The material is designed to be used in a realistic lighting environment, i.e. using full direct and indirect illumination.

In mental ray there are two basic methods to generate indirect light: Final Gathering and Global Illumination. For best results at least one of these methods should be used.

Traditional computer graphics light sources live in a cartoon universe where the intensity of the light doesn't change with the distance. The real world doesn't agree with that simplification. Light decays when leaving a light source due to the fact that light rays diverge from their source and the "density" of the light rays change over distance. This decay of a point light source is This decay of a point light source is 1/d², i.e. light intensity , i.e. light intensity is proportional to the inverse of the square of the distance to the source.

From a usage perspective, the shading model consists of three components:

• Diffuse - diffuse channel (including Oren Nayar "roughness").
• Reflections - glossy anisotropic reflections (and highlights).
• Refractions - glossy anisotropic transparency (and translucency).

The mia_material shading model

Direct and indirect light from the scene both cause diffuse reflections as well as translucency effects. Direct light sources also cause traditional "highlights" (specular highlights).

Raytracing is used to create reflective and refractive effects, and advanced importance-driven multi-sampling is used to create glossy reflections and refractions.

The rendering speed of the glossy reflections/refractions can further be enhanced by interpolation as well as "emulated" reflections with the help of Final Gathering.

The reflectivity of the wooden floor depends on the view angle

Many materials exhibit this behavior. Glass, water and other dielectric materials with fresnel effects (where the angular dependency is guided strictly by the Index of Refraction) are the most obvious examples, but other layered materials such as lacquered wood, plastic, etc. display similar characteristics.

The mia_material allows this effect both to be defined by the Index of Refraction, and also allows an explicit setting for the two reflectivity values for:

• 0 degree faces (surfaces directly facing the camera)
• 90 degree faces (surfaces 90 degrees to the camera)

The final surface reflectivity is in reality caused by the sum of three components:

• The Diffuse effect
• The actual reflections
• Specular highlights that simulate the reflection of light sources

Diffuse, Reflections and Highlights

In the real world "highlights" are just (glossy) reflections of the light sources. In computer graphics it's more efficient to treat these separately. However, to maintain physical accuracy the material automatically keeps "highlight" intensity, glossiness, anisotropy etc. in sync with the intensity, glossiness and anisotropy of reflections, hence there are no separate controls for these as both are driven by the reflectivity settings.

The transparency/translucency can treat objects either as solid or thin walled.

If all objects were treated as solids at all times, every single window pane in an architectural model would have to be modeled as two faces; an entry surface (that refracts the light slightly in one direction), and immediately following it an exit surface (where the light would be refracted back into the original direction).

Not only is this additional modeling work, it is a waste of rendering power to model a refraction that has very little net effect on the image. Hence the material allows modeling the entire window pane as one single flat plane, foregoing any actual "`refraction" of light.

Solid vs. Thin-walled transparency and translucency

In the above image the helicopter canopy, the window pane, the translucent curtain and the right sphere all use "thin walled" transparency or translucency, whereas the glass goblet, the plastic horse and the left sphere all use "solid" transparency or translucency.

Beyond the "physical" transparency (which models an actual property of the material) there is a completely separate non-physical "cutout opacity" channel to allow "billboard" objects such as trees, or to cut out things like a chain-link fence with an opacity mask.

Ambient Occlusion (henceforth referred to as "AO") is a method spearheaded by the film industry to emulate the "look" of true global illumination by using shaders that calculate how occluded (i.e. blocked) an area is from receiving incoming light.

Finally the mia_material contains a large set of built in functions for top performance, including but not limited to:

• Advanced importance sampling with ray rejection thresholds
• Adaptive glossy sample count
• Interpolated glossy reflection/refraction with detail enhancements
• Ultra fast emulated glossy reflections (refl_hl_only mode)
• Possibility to ignore internal reflections for glass objects
• Allowing a choice between traditional transparent shadows (suitable for e.g. a window pane) and refractive caustics (suitable for solid glass objects) on a per material basis.

diffuse_weight sets the desired level (and diffuse the color) of the diffuse reflectivity. Since the material is energy conserving, the actual diffuse level used depends on the reflectivity and transparency as discussed above.

The diffuse component uses the Oren-Nayar shading model. When diffuse_roughness is 0.0 this is identical to classical Lambertian shading, but with higher values the surface gets a a more "powdery" look:

Roughness 0.0 (left), 0.5 (middle) and 1.0 (right)

The reflectivity and refl_color together define level of reflections as well as the intensity of the traditional "highlight" (also known as "specular highlight").

Glossy reflections need to trace multiple rays to yield a smooth result, which can become a performance issue. For this reason there are a couple of special features designed to enhance their performance.

Translucency is handled as a special case of transparency, i.e. to use translucency there must first exist some level of transparency, and the refr_trans_w parameter decides how much of this is used as transparency and how much is translucency:

A transparency of 0.75 and a refr_trans_w of 0.0 (left), 0.5 (center) and 1.0 (right)

• If refr_trans_w is 0.0, all of the transparency is used for transparency.
• If refr_trans_w is 0.5, half of the transparency is used for transparency and half is used for translucency.
• If refr_trans_w is 1.0, all of the transparency is used for translucency and there is no actual transparency.

The translucency is primarily intended to be used in "thin walled" mode (as in the example above) to model things like curtains, rice paper, or such effects. In "thin walled" mode it simply allows the shading of the reverse side of the object to "bleed through".

The shader also operates in "Solid" mode, but the implementation of translucency in the mia_material is a simplification concerned solely with the transport of light from the back of an object to it's front faces and is not "true" SSS (sub surface scattering). An "SSS-like" effect can be generated by using glossy transparency coupled with translucency but it is neither as fast nor as powerful as the dedicated SSS shaders.

Solid translucency w. refr_trans_w of 0.0 (left), 0.5 (center) and 1.0 (right)

Anisotropic reflections and refractions can be created using the anisotropy parameter. The parameter sets the ratio between the "width" and the "height" of the highlights, hence when anisotropy is 1.0 there is no anisotropy, i.e. the effect is disabled.

For other values of anisotropy (above and below 1.0 are both valid) the "shape" of the highlight (as well as the appearance of reflections) change.

anisotropy values of 1.0 (left), 4.0 (center) and 8.0 (right)

The anisotropy can be rotated by using the anisotropy_rotation parameter. The value 0.0 is un-rotated, and the value 1.0 is one full revolution (i.e. 360 degrees). This is to aid using a texture map to steer the angle:

anisotropy_rotation values of 0.0 (left), 0.25 (center) and textured (right)

Note: When using a textured anisotropy_rotation it is important that this texture is not anti-aliased (filtered). Otherwise the anti-aliased pixels will cause local vortices in the anisotropy that appear as seam artifacts.

Built-in Ambient Occlusion

These parameters define some performance boosting options for reflections.

refl_falloff_dist allows limiting reflections to a certain distance, which both speeds up rendering as well as avoiding pulling in distant objects into extremely glossy reflections.

If refl_falloff_color is enabled and used, reflections will fade to this color. If it is not enabled, reflections will fade to the environment color. The former tends to be more useful for indoor scenes, the latter for outdoor scenes.

Full reflections (left), fading over 100mm (center) or 25mm (right)

Each material can locally set a maximum trace depth using the refl_depth parameter. When this trace depth is reached the material will behave as if the refl_hl_only switch was enabled, i.e. only show highlights and "emulated" reflections. If refl_depth is zero, the global trace depth is used.

refl_cutoff is a threshold at which reflections are rejected (not traced). It's a relative value, i.e. the default of 0.01 means that rays that contribute less than 1 to the final pixel are ignored.

The optimization settings for refractions (transparency) are nearly identical to those for reflections. The exception is that of refr_falloff_color which behaves differently.

• When refr_falloff_dist is used, and refr_falloff_color is not used, transparency rays will fade to black. This is like smoked glass or highly absorbent materials. Transparency will just completely stop at a certain distance. This has the same performance advantage as using the refl_falloff_dist for reflections, i.e. tracing shorter rays are much faster.

• However, when refr_falloff_color is used, it works differently. The material will then make physically correct absorption. Exactly at the distance given by refr_falloff_dist will the refractions have the color given by refr_falloff_color - but the rays are not limited in reach. At twice the distance, the influence of refr_falloff_color is double, at half the distance half, etc.

No limit (left), fade to black (center), fade to blue (right)

The leftmost cup has no fading. The center cup has refr_falloff_color off, and hence fades to black, which also includes the same performance benefits of limiting the trace distance as when used for reflections.

The options contain several on/off switches that control some of the deepest details of the material:

Glossy reflections and refractions can be interpolated. This means they render faster and become smoother.

Interpolation works by pre-calculating glossy reflection in a grid across the image. The number of samples (rays) taken at each point is govern by the refl_samples or refr_samples parameters just as in the non-interpolated case. The resolution of this grid is set by the intr_grid_density parameter.

However, interpolation can cause artifacts. Since it is done on a low resolution grid, it can lose details. Since it blends neighbors of this low resolution grid it can cause over-smoothing. For this reason it is primarily useful on flat surfaces. Wavy, highly detailed surfaces, or surfaces using bump maps will not work well with interpolation.

Valid values for intr_grid_density parameter is:

• 0 = grid resolution is double that of the rendering
• 1 = grid resolution is same as that of the rendering
• 2 = grid resolution is half of that of the rendering
• 3 = grid resolution is a third of that of the rendering.
• 4 = grid resolution is a fourth of that of the rendering.
• 5 = grid resolution is a fifth of that of the rendering.

Within the grid data is stored and shared across the points. Lower grid resolutions is faster but lose more detail information. Both reflection and refraction has an intr_refl_samples parameter which defines how many stored grid points (in an N by N group around the currently rendered point) is looked up to smooth out the glossiness. The default is 2, and higher values will "smear" the glossiness more, but are hence prone to more overmoothing artifacts.

No interpolation (left), looking up 2 points (center) and 4 points (right)

The reflection of the left cup in the floor is not using interpolation, and one can perceive some grain (here intentionally exaggerated). The floor tiles under the other two cup uses a half resolution interpolation with 2 (center) and 4 (right) point lookup respectively.

This image also illustrates one of the consequences of using interpolation: The foot of the left cup, which is near the floor, is reflected quite sharply, and only parts of the cup far from the floor are blurry. Whereas the interpolated reflections on the right cups have a certain "base level" of blurriness (due to the smoothing of interpolation) which makes even the closest parts somewhat blurry. In most scenes with weak glossy reflections this discrepancy will never be noticed, but in other cases this can make things like legs of tables and chairs feel "unconnected" with a glossy floor, if the reflectivity is high.

To solve this the intr_refl_ddist parameter exists. It allows a second set of detail rays to be traced to create a "clearer" version of objects within that radius.

No detail distance (left), 25mm detail distance (center) and 150mm detail distance (right)

All three floor tiles use interpolation but the rightmost two use different distances for the "detail distance".

This also allows an interesting "trick": Set the refl_samples to 0, which renders reflections as if they were mirror-perfect but use the interpolation to introduce blur into this "perfect" reflection (and perhaps use the intr_refl_ddist to make nearby parts less blurry). This is an extremely fast way to obtain a glossy reflection.

No detail distance (left), with detail distance (right)

The above floor tiles are rendered with mirror reflections, and the "blurriness" comes solely from the interpolation. This renders as fast (or faster!) than pure mirror reflections, yet gives a satisfying illusion of true glossy reflections, especially when utilizing the intr_refl_ddist as on the right.

The mia_material also supports the following special inputs:

The bump parameter accepts a shader that perturbs the normal for bump mapping. This parameter is only used it the new bump_mode parameter is zero.

When no_diffuse_bump is off, the bumps apply to all shading components (diffuse, highlights, reflections, refractions... ). When it is on, bumps are applied to all component except the diffuse. This means bumps are seen in reflections, highlights, etc. but the diffuse shading shows no bumps. It is as if the materials diffuse surface is smooth, but covered by a bumpy lacquer coating.

no_diffuse_bump is off (left) and on (right)

In mia_material_x there are also three new parameters related to bump mapping: two vector bump inputs, overall_bump and standard_bump, and a bump_mode parameter defining the coordinate-space of those vectors. The shaders put into overall_bump or standard_bump should return a vector, but it is also legal for those shaders to modify the normal vector themselves and return (0,0,0).

overall_bump defines an overall bump that always applies both to the diffuse and the specular component at all times, regardless of the setting of no_diffuse_bump standard_bump is a vector equivalent of the old bump parameter, in that it applies globally when no_diffuse_bump is off, and only to the specular/reflection "layer" when no_diffuse_bump is on. However, the standard_bump is added "on top of" the overall_bump result.

The intended use is to put the mia_roundcorners shader in overall_bump and your normal bump shader into standard_bump. This way, the "round corners" effect will apply both to the diffuse and specular component irregardless of the setting of no_diffuse_bump.

The "add" modes mean that the vector should contain a normal perturbation, i.e. a modification that is "added" to the current normal. Whereas "set" mode means that the actual normal is replaced by the incoming vector, interpreted in the aforementioned coordinate space.

This new scheme makes the mia_material_x bump mapping compatible with more mental ray integrations, as well as allows the round corners to be applied even if no_diffuse_bump is on.

The cutout_opacity is used to apply an opacity map to completely remove parts of objects. A classic example is to map an image of a tree to a flat plane and use opacity to cut away the parts of the tree that are not there.

Mapping the transparency (left) vs. cutout_opacity (right)

The additional_color is an input to which one can apply any shader. The output of this shader is simply added on top of the shading done by the mia_material and can be used both for "self illumination" type effects as well as adding whatever additional shading one may want.

The material also supports standard displacement and environment shaders. If no environment is supplied, the global camera environment is used.

Here follows a detailed listing of the available outputs of mia_material_x:

Most of the outputs follow the pattern of xxx_result, xxx_raw and xxx_level. The "result" is the final contribution, "raw" is the un-scaled contribution, and "level" is the scaling. The "level" is often related to an input parameter (or combinations thereof), and has been modified to abide by the energy conservation feature of the material.

Unless otherwise noted, it is true that Unless otherwise noted, it is true that xxx_result = xxx_raw * xxx_level. .

The different outputs and their relationship

Hence the outputs contain some redundancy; if one just wants the "current reflections" in a separate channel, use refl_result, but if one wants more control over the amount of reflections in post production, one can instead use refl_raw and refl_level, multiplying them in the compositing phase prior to adding them to the final color.

Be aware, though, that mia_material_x will intentionally sample reflections that has a very low level in the actual rendering phase at low quality (for performance), so doing huge modifications to reflection intensity in post should be avoided.

The following outputs exist:

Due to the redundancy available in the outputs, there are several ways to composite them together to yield the same result as the beauty render. Here we outline two compositing pipelines in equation form.

First we have the "simple" variant, which is simply a sum of the various result parameters. This version allows only minor post production changes to the overall balance between the materials.

But it has the advantage of not needing as many files, as well as working reasonably well in non-floating-point compositing.

    Beauty = diffuse_result + indirect_result + spec_result + 
                refl_result + refr_result + tran_result + 
                add_result

Then we have the more "complex" variant which uses the various raw and level outputs, which allows much greater control in post production.

Note that the raw outputs needs to be stored and composited in floating point to maintain the dynamic range. The level outputs always stay in a 0.0-1.0 range and does not require floating point storage.

    Beauty = diffuse_level * (diffuse_raw + (indirect_raw * ao_raw)) + 
                spec_level * spec_raw +
                refl_level * refl_raw +
                refr_level * refr_raw +
                tran_level * tran_raw +
                add_result

The Final Gathering algorithm in mental ray 3.5 is vastly improved from earlier versions, especially in it's adaptivity. This means one can often use much lower ray counts and much lower densities than in previous versions of mental ray.

Here are some quick rules-of-thumb for creating various materials. They each assume basic default settings as a starting point.

This is the kind of "hybrid" materials one run into in many architectural renderings; lacquered wood, linoleum, etc.

For these materials brdf_fresnel should be off (i.e. we define a custom BRDF curve). Start out with brdf_0_degree_refl of 0.2 amd brdf_90_degree_refl of 1.0 and apply some suitable texture map to the diffuse. Set reflectivity around 0.5 to 0.8.

How glossy is the material? Is reflections very clear or very blurry? Are they Strong or Weak?

• For clear, fairly strong reflections, keep refl_gloss at 1.0

• For slightly blurry but strong reflections, set a lower refl_gloss value. If performance becomes an issue try using refl_interpolate.

• For slightly blurry but also very weak reflections one can often "cheat" by setting a lower refl_gloss value (to get the broader highlights) but set refl_samples value to 0. This shoots only one mirror ray for reflections - but if they are very weak, one can often not really tell.

• For medium blurry surfaces set an even lower refl_gloss and maybe increase the refl_samples. Again, for performance try refl_interpolate.

• For extremely blurry surfaces or surfaces with very weak reflections, try using the refl_hl_only mode.

A typical wooden floor could use refl_gloss of 0.5, refl_samples of 16, reflectivity of 0.75, a nice wood texture for diffuse, perhaps a slight bump map (try the no_diffuse_bump checkbox if bumpiness should appear only in the lacquer layer).

A linoleum carpet could use the same but with a different texture and bump map, and probably with a slightly lower reflectivity. and refl_gloss.

Ceramic materials are glazed, i.e. covered in a thin layer of transparent material. They follow similar rules to the general materials mentioned above, but one should have brdf_fresnel on and the refr_ior set at about 1.4 and reflectivity at 1.0.

The diffuse should be set to a suitable texture or color, i.e. white for white bathroom tiles.

Stone is usually fairly matte, or has reflections that are so blurry they are nearly diffuse. The "powdery" character of stone is simulated with the diffuse_roughness parameter - try 0.5 as a starting point. Porous stone such as bricks would have a higher value.

Stone would have a very low refl_gloss (lower than 0.25) and one can most likely use refl_hl_only to good effect for very good performance. Use a nice stone texture for diffuse, some kind of bump map, and perhaps a map that varies the refl_gloss value.

The reflectivity would be around 0.5-0.6 with brdf_fresnel off and brdf_0_degree_refl at 0.2 and brdf_90_degree_refl at 1.0

Glass is a dielectric, so brdf_fresnel should definitely be on. The IOR of glass is around 1.5. Set diffuse_weight to 0.0, reflectivity to 1.0 and transparency to 1.0. This is enough to create basic, completely clear refractive glass.

If this glass is for a window pane, set thin_walled to on. If this is a solid glass block, set thin_walled to off and consider if caustics are necessary or not, and set do_refractive_caustics accordingly.

Is the glass frosted? Set refr_gloss to a suitable value. Tune the refr_samples for good quality or use refr_interpolate for performance.

For clear glass the tips in the previous section work. But colored glass is a slightly different story.

Many shaders set the transparency at the surface of the glass. And indeed this is what happens if one simply sets a refr_color to some value, e.g. blue. For glass done with thin_walled turned on this works perfectly. But for solid glass objects this is not an accurate representation of reality.

Water, like glass, is a dielectric with the IOR of 1.33. Hence, the same principles as for glass (above) applies for solid bodies of water which truly need to refract things... for example water running out of a tap. Colored beverages use the same principles as colored glass, etc.

Water into Wine

To create a beverage in a container as in the image above, it is important to understand how the mia_material handles refraction through multiple surfaces vs. how the "real world" tackles the same issue.

What is important for refraction is how the transition from one medium to another with a different IOR. Such a transition is known as an interface.

For lemonade in a glass, imagine a ray of light travelling through the air (IOR = 1.0) enter the glass, and is refracted by the IOR of the glass (1.5). After travelling through the glass the ray leaves the glass and enters the liquid, i.e. it passes an interface from one medium of IOR 1.5 to another medium of IOR 1.33.

One way to model this in computer graphics is to make the glass one separate closed surface, with the normals pointing towards the inside of the glass and an IOR of 1.5, and a second, closed surface for the beverage, with the normals pointing inwards and an IOR of 1.33, and leaving a small "air gap" between the container and the liquid.

While this "works", there is one problem with this approach: When light goes from a higher IOR to a lower there is a chance of an effect known as "Total Internal Reflection" (TIR). This is the effect one sees when diving in a swimming pool and looking up - the objects above the surface can only be seen in a small circle straight above, anything below a certain angle only shows a reflection of the pool and things below the surface. The larger the difference in the IOR of the two media, the larger is the chance of TIR.

So in our example, as the ray goes from glass (IOR=1.5) to air, there is a large chance of TIR. But in reality the ray would move from a medium of IOR=1.5 to one of IOR=1.33, which is a much smaller step with a much smaller chance of TIR. This will look different:

Correct refraction (left) vs. the "air gap" method (right)

The result on the left is the correct result, but how it is obtained?

The solution to the problem is to rethink the modeling, and not think in terms of media, but in terms of interfaces. In our example, we have three different interfaces, where we can consider the IOR as the ratio between the IOR's of the outside and inside media:

• Air-Glass interface (IOR = 1.5/1.0 = 1.5)
• Air-Liquid interface (IOR = 1.33/1.0 = 1.33)
• Glass-Liquid interface (IOR=1.33/1.5=0.8)

It is evident that in the most common case of an interface with air, the IOR to use is the IOR of the media (since the IOR of air is 1.0), whereas in an interface between two different media, the situation is different.

To correctly model this scenario, we then need three surfaces, each with a separate mia_material applied:

The three interfaces for a liquid in a glass

• The Air-glass surface (blue), with normals pointing out of the glass, covering the area where air directly touches the glass, having an IOR of 1.5
• The Air-liquid surface (green), with normals pointing out of the liquid, covering the area where air directly touches the liquid, having an IOR of 1.33
• The Glass-liquid surface (red), with normals pointing out of the liquid, covering the area where the glass touches the liquid, having an IOR of 0.8

By setting a suitable refr_falloff_dist and refr_falloff_color for the two liquid materials (to get a colored liquid), the image on the left in the comparison above is the result.

A water surface is a slightly different matter than a visibly transparent liquid.

The ocean isn't blue - it is reflective. Not much of the light that goes down under the surface of the ocean gets anywhere of interest. A little bit of it is scattered back up again doing a little bit of very literal "sub surface scattering".

To make an ocean surface with the mia_material do the following steps:

Set diffuse_weight to 0.0, reflectivity to 1.0 and transparency to 0.0 (yes, we do not use refraction at all!).

Set the refr_ior to 1.33 and brdf_fresnel to on. Apply some interesting wobbly shader to bump and our ocean is basically done!

This ocean has only reflections guided by the IOR. But this might work fine - try it. Just make sure there is something there for it to reflect! Add a sky map, objects, or a just a blue gradient background. There must be something or it will be completely black.

The Ocean isn't blue - the sky is

For a more "tropical" look, try setting diffuse to some slight greenish/blueish color, set the diffuse_weight to some fairly low number (0.1) and check the no_diffuse_bump checkbox.

Now we have a "`base color" in the water which emulates the little bit of scattering occurring in the top level of the ocean.

Enjoy the tropics

Brushed metal is an interesting special case of metals. In some cases, creating a brushed metal only takes turning down the refl_gloss to a level where one receives a "very blurred" reflection. This is sufficient when the brushing direction is random or the brushes are too small to be visible even as an aggregate effect.

For materials that have a clear brushing direction and/or where the actual brush strokes are visible, creating a convincing look is a slightly more involved process.

The tiny grooves in a brushed metal all work together to cause anisotropic reflections. This can be illustrated by the following schematic, which simulates the brush grooves by actually modeling many tiny adjacent cylinders, shaded with a simple Phong shader:

Many small adjacent cylinders

As one can see, the specular highlight in each of the cylinders work together to create an aggregate effect which is the anisotropic highlight.

Also note that the highlight isn't continuous, it is actually broken up in small adjacent segments. I.e. the main visual cues that a material is "brushed metal' are:

• Anisotropic highlights that stretch out in a direction perpendicular to the brushing direction.
• A discontinuous highlight with "breaks" in the brushing direction.

Many attempts to simulate brushed metals have only looked at the first effect, the anisotropy. Another common mistake is to think that the highlight stretches in the brushing direction. Neither is true.

Hence, to simulate brushed metals, we need to simulate these two visual cues. The first one is simple; use anisotropy and anisotropy_rotation to make anisotropic highlights. The second can be done in several ways:

• With a bump map
• With a map that varies the anisotropy or refl_gloss
• With a map that varies the refl_color

Each have advantages and disadvantages, but the one we will try here is the last one. The reason for choosing this method is that it works well together with interpolation.

1. Create a map for the "brush streaks". There are many ways to do this, either by painting a map in a paint program, or by using a "Noise" map that has been stretched heavily in one direction.

2. The map should vary between middle-gray and white. Apply this map to the refl_color in a scale suitable for the brushing.

3. Set diffuse to white (or the color of the metal) but set diffuse_weight to 0.0 (or a small value).

4. Make sure refl_is_metal is enabled.

5. Set refl_gloss to 0.75.

6. Set anisotropy to 0.1 or similar. Use anisotropy_rotation to align the highlight properly with the map. If necessary use anisotropy_channel to base it on the same texture space as the map.

Brushed Metal