Shadowing Reference

Document Summary: A reference detailing the shadowing methods used in Unreal Engine 3.

Overview

Unreal Engine 3 uses several distinct shadowing solutions:

• Whole-scene dynamic shadows - Movable lights use whole-scene dynamic shadows.
• Per-object dynamic shadows - Immovable lights illuminating movable or dynamic objects use per-object dynamic shadows.
• Precomputed shadows - Immovable lights illuminating static objects use precomputed shadow maps.
• Precomputed lighting - Static lights illuminating static objects use precomputed light maps.

Light classes

For each type of light you can place in the world, there are different classes that control the shadowing behavior of the light. Those classes limit the light from changing in certain ways at runtime, or designate the light as a "dominant" light.

• The base class of each light (e.g. PointLight) is the static version of that light. It uses per-object dynamic shadows combined with precomputed lighting.
• The Movable subclass of that light may be moved at runtime. It uses whole-scene dynamic shadows.
• The Toggleable subclass of that light may not be moved, but it can change color and brightness at runtime. It uses per-object dynamic shadows combined with shadow maps.
• The Dominant subclass of that light is similar to Toggleable lights, but uses distance field shadow maps.

SpotLight	Movable SpotLight	Toggleable SpotLight	Dominant SpotLight

Recommended usage

For most lights, you should use the base light classes: PointLight, SpotLight and DirectionalLight. These classes have the least impact on runtime performance, since they all use precomputed lighting.

For lights that don't need to move, but need to change their color, brightness or enabled state at runtime, you should use the Toggleable light subclasses: PointLightToggleable, SpotLightToggleable, DirectionalLightToggleable. These classes use precomputed shadows together with per-pixel lighting at runtime, and only have a small impact on runtime performance.

For most scenes you will want to have a small number of lights that uses precomputed shadows with higher quality lighting than you can get with precomputed lighting. For that, you would use the Dominant light subclasses: DominantPointLight, DominantSpotLight and DominantDirectionalLight. These classes are similar to Toggleable light subclasses, but are restricted so each object can only be affected by a single Dominant light. This allows Dominant lights to be slightly more efficient at runtime than Toggleable lights. For more information, see Dominant Lights.

For cases where you need a light to move, you must use the Movable light subclasses: PointLightMovable, SpotLightMovable, and DirectionalLightMovable. These classes compute shadowing and lighting every frame at runtime, so they have the largest impact on runtime performance.

Shadow types

Per-object dynamic shadows

Immovable lights will use per-object dynamic shadows for the dynamic objects that they affect. The per-object dynamic shadows work by rendering a depth buffer for the object from the perspective of the light and then projecting it onto the world. Pixels that are further from the light than the depth in the projected shadow depth buffer are shadowed. A filter is used to create a soft edge between shadowed and lit areas.

In this screen shot, the flying vehicle (Cicada) is casting a soft dynamic shadow on the bridge.

The per-object dynamic shadow's quality and cost depends on how large the projected depth buffer frustum is on the screen, which depends on how big the dynamic object's bounding box is. This means that it works best on objects with a tight bounding box. Skeletal meshes without an associated physics asset will have a loose bounding box that doesn't change at runtime depending on joint animation, so they will have a low quality and expensive shadow. Thus skeletal meshes with a physics asset will have a tight bounding box computed each frame from the joint animation, so they will have higher quality and cheaper shadows. For that reason, you should create a physics asset^? for every skeletal mesh, regardless of whether it will have physics-based animation or not.

You can use the "show shadowfrustums" console command, or the shadow frustums option in the editor viewport's Show->Advanced Flags menu, to view the frustums used for per-object dynamic shadows.

In this screen shot, the skeletal mesh does not have a physics asset assigned to it. Thus it's shadow frustum which is represented by the blue box (black outline added for visibility) is large and inefficient.

In this screen shot, the skeletal mesh has a physics asset assigned to it. Thus it's shadow frustum which is represented by the blue box (black outline added for visibility) is tight and efficient.

To handle shadows from static objects cast on dynamic objects, non-movable lights will create a second per-object projected shadow depth buffer. This depth buffer will be created from the static objects between the light and the dynamic object.

In this screen shot, this skeletal mesh has shadows casted on it from precomputed shadows. Ensure that you have a DynamicLightEnvironments assigned to the dynamic skeletal mesh, with "Use Boolean Environment Shadowing" unchecked to allow dynamic pre-shadowing (off by default for SkeletalMeshCinematicActor).

In this screen shot, this interpolated static mesh has shadows casted on it from the static environment geometry. Ensure that you have a DynamicLightEnvironment assigned to the interpolated static mesh, with "Use Boolean Environment Shadowing" unchecked to allow dynamic pre-shadowing (on by default for InterpActor).

Modulated shadows

Per-object dynamic shadows may be modulated or normal. Normal shadows mask the contribution of the shadow-casting light to the shadow receiver, while modulated shadows only modulate the shadow receiver's final color by the modulated shadow color. Static lights require modulated shadows to combine the per-object dynamic shadows with the precomputed lighting that such static lights use.

In this screen shot, normal shadowing is applied.

In this screen shot, modulated shadowing is applied with Mod Shadow Color set to 0.f, 0.f, 0.f, 1.f (Black).

In this screen shot, Mod Shadow Color is set to 1.f, 0.f, 0.f, 1.f (Red).

Static lights and toggleable lights default to modulated shadows, while dominant lights default to normal shadows.

Whole-scene dynamic shadows

For Movable lights, whole-scene dynamic shadows are used. The whole-scene shadows work similarly to the projected shadow depth buffers of per-object dynamic shadows, but instead of creating a projected shadow-depth buffer for each object, they use a fixed number of shadow depth buffers for the whole scene. A spot light uses a single shadow depth buffer, point lights use six shadow depth buffers corresponding to the faces of a cube, and directional lights use one or more cascaded shadow depth buffers centered around the viewer.

In this screen shot, a movable point light is casting whole-scene shadows.

In this screen shot, a movable spot light is casting whole-scene shadows.

In this screen shot, a moveable dominant directional light is casting whole-scene shadows.

Unlike per-object dynamic shadows, whole-scene dynamic shadows cannot be modulated.

Cascaded shadow maps

Dominant directional lights can use cascaded shadow depth buffers for whole-scene shadows. If the light is immovable, the cascaded shadow depth buffers can be blended with the precomputed shadows, using cascaded shadow depth buffers for nearby objects and precomputed shadows for distant objects.

Precomputed shadows

Toggleable lights use precomputed shadows. The precomputed shadows are stored using either a gray scale mask that is applied when rendering the light, or for dominant lights as a signed distance field.

Precomputed lighting

Static lights use precomputed lighting on static objects. The precomputed lighting combines all static lights together, so the lights can no longer be separated at runtime. Because of that, static lights can only cast modulated per-object shadows from dynamic objects.

Shadow opacity

Masked material opacity will be respected by all types of shadows.

Translucent material opacity will only be respected for shadows cast by static objects from Toggleable lights or static lights.

To make the translucent materials cast dynamic shadows as masked materials, set the material's Cast Lit Translucency Shadow As Masked property to true and set the Opacity Mask material parameter.

Shadow and light-map resolution

Precomputed Shadows and BSP

To change the resolution of the shadow maps and light maps used for a BSP surface, select the surface or surfaces, and hit F5. This will bring up the Surface Properties menu.

Under Pan/Rot/Scale you will find a Light map value. The lower the value, the higher resolution the resulting shadows will be, as illustrated in the image below:

Precomputed Shadows and Static Meshes

Precomputed shadows are cast on static mesh geometry as well. By default, shadow factors are calculated per vertex. For better shadow resolution, a shadow map should be generated for each static mesh.

A shadow map requires that all UV data be unique. No mirroring or tiling is allowed. If your static mesh's UVs have mirroring, stacking, or tiling, then you will need to create a second set of new unique UVs. For more information about how to set up multiple sets of UVs see the VertexBlendingTutorial.

Open your static mesh in the Static Mesh Editor (double click it in the Generic Browser window). Fill the desired value into LightMapResolution. The default value of 0 means no shadow map will be created.

Keep in mind that an independent shadow map will be generated for each instance of your mesh in the level file, which could potentially have dramatic memory ramifications.

If your mesh has multiple UVs, select the set you want to use for the shadow map in LightMapCoordinateIndex. The default is 0, which is the original set of UVs.

Notice that the number of UV channels is listed in the top left of the render window.

Below is an illustration of per-vertex shadowing vs. shadow maps.

Shadowing options