Hand-authoring LPE can be a tedious process and for artists it's not always desired to do these manually. This document aims to improve the artist's understanding and usage of LPE. Understanding these concepts and applying them will make you a hero to many pipelines given the power and flexibility of the feature. |
In production it often becomes necessary to create non-physical effects and art-directed changes to your shot. Light Path Expressions (LPE) are a powerful way to collect and output specific light paths for alteration later.
Some examples of the power of LPE are the ability to:
Note that trace sets can also be used to accomplish some of these effects but since trace sets are global in the render, LPE will obey these trace sets (meaning data that would have been collected otherwise is restricted by the trace set). Light Path Expressions handle light transport, as such they collect illumination by default and not shadows (shadows are simply a lack of light) A specialized Shadow LPE is provided for compositing tasks as part of our Holdout workflow. Since these follow the physics of light, your LPE will use addition (plus) and subtraction (minus) as adjustments in compositing. This avoids artifacts introduced by multiplication and division that should happen at the sample level, not on final pixels. |
LPE, Cryptomatte, and Trace Sets should improve the flexibility of your pipeline and when managed correctly, avoid expensive re-rendering for artistic tweaks.
An excellent primer with useful visuals can be found in a tutorial video from Lollipop Shaders. Begin at Class 6 for the LPE introduction. |
To understand what is collected in an LPE, we will talk about the parts below and some context. Firstly, we always begin at the camera, so we begin an expression using C for Camera as the starting point for light collection.
Light is scattered when it strikes an object. It is either reflected off, absorbed, transmitted, or some combination of the three. In the LPE syntax we list two types of events:
The tokens for these events are are R and T as shown above. You may specify which you want to collect or both. Later we will discuss some shorthand for specifying either but for now we'll use the tokens. Light that is absorbed is handled by the material itself and since it doesn't travel back into the scene (its energy absorbed like light in the real world) we don't collect it.
Below is a beauty render and the diffuse and specular transmission effects rendered alone in an LPE. This includes translucency and subsurface scattering.
Diffuse Transmission: lpe:C<TD>*[<L.>O] Specular Transmission: lpe:C<TS>*[<L.>O] |
Below is a beauty render and the diffuse and specular reflection effects rendered alone in an LPE.
Diffuse Reflection: lpe:C<RD>*[<L.>O] Specular Reflection: lpe:C<RS>*[<L.>O] |
A scattering event is light scattered from a reflection or transmission. In RenderMan we provide two events and one special User event:
You can specify a chain of events if you like, such as DDS, meaning Diffuse event to Diffuse event to a Specular in that order. The LPE would collect anything that happens in this specific order and store them. For simplicity we'll cover more common cases first.
Diffuse (All types, direct and indirect): lpe:C<[RT]D>*[<L.>O] Specular (All types, direct and indirect): lpe:C<[RT]S>*[<L.>O] |
In ray tracing images, we mimic what happens as light travels through a real world scene.
There is a funny thing about specular objects without diffuse shading, they only look like the things they reflect. If you were to render the Cornell Box scene examples with perfectly specular surfaces and a light, all you would see are reflections of the light. What does a mirror look like? Can you describe a mirror that has nothing to reflect? Or does the mirror look like what it's reflecting? This is why indirect results are so important for realism and ray tracing.
A diffuse scene using subsurface and translucent (both called diffuse transmission) effects.
A specular scene using iridescence and transmission (glass).
Now that we have some understanding of the main components, let's begin by looking at the parts used to organize an LPE
< > are used to choose/define a single event. For example, when deciding I want a specular reflection event, I can say <RS> for a reflective specular event. I cannot create an event that is either/or, meaning I can't say <RT> because that's two different types (reflection and transmission). Instead it must be a single type and event defined in the angle brackets. You can also put these in order to collect a light path in a specific order, such as <RS ><TD > as to say "One reflective specular event and then a transmissive diffuse event next".
[ ] are used to define events of any/either/or. So in the example above I could say [RT] meaning one Reflection or Transmission type and [DS] which means Diffuse or Specular event. This is pretty common when you need to collect light along a path with many different possibilities. An example might be <[RT][DS]> which means a single event (note the angle brackets) with either a reflective or transmissive type (first set of square brackets) of a diffuse or specular event (the second set of square brackets).
() are used to encapsulate a series of events. This operates most like the purpose of parenthesis in a mathematical operation, grouping items to be considered together. Such an example would be to have two different sets of events you want to collect but placing a pipe, | to say "or" between these encapsulated events. C(DS)|(TD)L would be Camera to Diffuse to Specular event or Camera to Diffuse Transmission event.
Above we are explicitly stating what we want using the tokens. We can use shorthand and some following examples will use it. I can substitute a . (period) for the events and types.
Instead of writing out the tokens, just use a . for "whatever" if you need not be specific.
Diffuse (All types, direct and indirect): lpe:C<.D>*[<L.>O] Specular (All types, direct and indirect): lpe:C<.S>*[<L.>O] |
You could even shorten the beauty AOV by saying:
A longer beauty LPE can be seen as Beauty: lpe:C<[RT][DS]>.*[<L.>O] Can be written as a shorter version below Beauty: lpe:C<..>*[LO] |
This can be shrunk even further (you notice we remove a couple things where "any" or the . just gets omitted) but we'll go over what the LPE parser assumes later. For now we'll keep it tidy and readable.
Since we're tracing through the scene collecting light, it might be important to specify how much light you need or want to collect. This is usually meant to separate the direct and indirect events mentioned above. But you may also specify how many bounces along this path you want to collect rather than let the renderer collect all of them. Collecting specific bounces is covered later.
To begin we'll look at the usual notation for Direct and then Indirect light collection.
Above we were defining events and their type. But we need a way to say how many of these we want. We could manually string together events in a particular order but that becomes a painful experience in copy and paste. It's better to let the renderer make the decision for you. We have two common ways to define how many of our defined events we want.
* means zero or more of these preceding events. Since it's zero or more, it will collect a direct lighting event and the ones that follow it (meaning indirect) til they reach a light source. It's the simplest way to capture light for the whole path. The following collects only diffuse reflection and all the events that match diffuse reflection until it strikes a light.
lpe:C<RD>*[<L.>O] |
+ means one or more events. This means after the initial event, begin collecting others. This means it will collect indirect lighting only as the first (direct) event is skipped. Below is only indirect diffuse reflection (the typical indirect light AOV)
lpe:C<RD>+[<L.>O] |
If none of the notations are used, it collects an event and then goes to the light, collecting only direct lighting. Below we omit the * and the +, only direct diffuse reflection is collected.
lpe:C<RD>[<L.>O] |
Above we use the most common ways to collect light paths, you either want all, direct, or indirect. This handles most scenarios you come across. But there may be reason to do something with a specific bounce or set of bounces. This can be anything from non-physical scaling of the light (brighter or darker) or isolating a specific bounce where noise is introduced and subtracting it from the final result. This last example is best when you can isolate the object(s) causing the noise. This will be covered under LPE Groups. Below is an example of rendering only the first (direct light), second, and third light bounce. These LPE are all very similar to the Beauty LPE but we exchange the * for all bounces with a specific bounce contained in a curly bracket {} For example, {1} collects the first bounce or the direct light frm the ceiling rect light to the object and then immediately to the camera. The second bounce is {2} which says the light must interact with something one more time before being stored, this is the first indirect bounce after light meets a surface. This goes on until the path length is terminated by the parameters in the Render Settings, Max Specular/Diffuse depth and/or Max Path Length for the integrator and possibly Russian Roulette early termination (all the energy was already absorbed leaving you with no more information to store).
lpe:C<[RT][DS]>{1}[<L.>O]
lpe:C<[RT][DS]>{2}[<L.>O]
lpe:C<[RT][DS]>{3}[<L.>O] |
Let's look at an expanded beauty LPE. And by "expanded" I mean with all the parts spelled out and not in shorthand.
lpe:C<[RT][DS]>*[<L.>O] |