User Manual for XNA Procedural LTrees
Overview of Classes
A tree is composed of three parts: A TreeSkeleton
and a TreeLeafCloud
. The skeleton describes the topology of the tree, and here you can inspect the positions of each individual branch and leaf (if you wanted to). The tree mesh contains
the geometry required to render the branches, but it must be given an
from outside to render itself -- it sets no effect parameters on its own. Likewise, the tree leaf cloud contains the vertex buffer for the leaves.
contains all you need to render a tree, including the skeleton, mesh, leaf cloud, textures, effects, and animation state. It will set effect parameters in its draw calls for you.
contains the rotations of each bone in a tree, and is stored independently of the other parts, so one tree can be used in several copies.
is the class that generates the random tree. It produces a tree skeleton, which can be used to create a tree mesh and a leaf cloud.
The tree and leaf shaders set a couple of render states automatically, but they do not change them back. Managing render states is a central aspect of any game, and LTrees will not do it for you. Therefore,
it is your own responsibility to restore the previous render states
. If you are having graphical issues with the trees, it is most likely a render state issue.
You need two objects: A WindSource
which describes the wind strength at different times and places, and the
is an interface so you can provide your own definition of wind strength, but a simple implementation is included in the project.
To get started, add these two fields in your Game class or similar location:
public class Game1 : Game
Create them during initialization:
protected override void LoadContent()
wind = new WindStrengthSin();
animator = new TreeWindAnimator(wind);
In the Update method, update the wind source so it knows what time it is, and then use it to animate your trees:
protected override void Update(GameTime gameTime)
protected override void Draw(GameTime gameTime)
foreach (SimpleTree tree in MyCollectionOfTrees)
animator.Animate(tree.Skeleton, tree.AnimationState, gameTime);
Note that the WindStrengthSin
was designed to demonstrate the wind feature. In practice, it can be tiresome when all the trees are blowing in such an irregular wind. Currently, you are encouraged to write your own implementation
, which also allows you to synchronize it with the wind that affects other scenery like grass and snowflakes.
- Caution: The BoundingSphere property of the TreeMesh is only guaranteed to enclose the tree in its initial animation state.
- Hint: Trees in your world can share the same mesh but still use different animation states, so they can be animated independently.
Using Multiple Levels-of-Detail (LOD)
A common way to improve performance is to reduce the amount of detail on models the further they are from the camera. When sufficiently far away, you want to display a billboard instead of rendering the tree, but that is another topic ("Impostering").
Before turning the tree into a billboard, we can decrease the number of polygons in the trunk.
To create a low-detail model of a tree, try this:
TreeMesh lowPolyMesh = new TreeMesh(GraphicsDevice, tree.Skeleton, 4);
The third argument is the number of radial segments to use in the root branch. The default is 8 -- a value of 4 makes it look like a square, but should hopefully not be noticeable at a distance. Note that the number of radial segments in branches will gradually
decrease towards 3 (the minimum) as the radius gets smaller.
There is currently no way to decrease the level of detail on leaves.
Using Features like Fog, Point Lights, and Shadows
The shaders provided in the project only allow for two directional lights, hardware-skinned animation, and nothing else. In modern games this is insufficient, but the exact requirements depends too strongly the individual application, which is why serious users
are encouraged to write their own shaders and/or a replacement for
to render their trees. The reference shaders are included in the content project bundled in LTreeLibrary.
To replace the default shaders inserted by the content loader, add this before
loading any trees:
TreeProfileReader.DefaultTreeShaderAssetName = "Effects/<YourShaderHere>";
This way, you can still use the TreeProfile
loaded by the content manager -- nobody says you have to call
to generate your trees.
has all the properties available required to make your own implementation.
Alternatively, you can change the effect loaded for each individual tree profile by settings its "Trunk Effect" and/or "Leaf Effect" properties in the property editor.
- Note: The tree mesh contains holes and is self-intersecting, so stencil shadows (aka. shadow volumes) cannot be used to draw shadow. Shadow mapping is the recommended approach.
Avoiding the Content Pipeline
For those who don't like the content pipeline, an
-file can be loaded at runtime using the static method
. Note that this method does not load textures or effects for you. Use the resulting
to generate a
, and from the skeleton create a TreeMesh
Creating more Tree Profiles
This is not an easy task. One way is of course to copy an existing tree profile and simple replace the texture with new ones. To change the texture used, open the .ltree file in your favourite XML editor (double-clicking it in visual studio will work). Find
these two lines:
Replace the inner text of these two tags with the asset names of the trunk texture and leaf texture you want.
To understand how the structure of the tree is generated, you must know what an L-system is (LTrees got its name from "L-system"). I suggest starting with an
interactive example in 2D
. The tree generator works like this, only in three dimensions. Think of it as moving around a crayon in 3D space, and this crayon paints tree branches in its wake. The L-system produces a long sequence
of instructions for the crayon. Instructions include to move forward, rotate around its own X or Y axis, scale, store the current state, restore the previously stored state, place a leaf here, place a bone here, and so on.
contains a sequence of instructions, and the instruction Call
makes the crayon execute the instructions in a specified production, much like a method call in C#. The
tag instructs the crayon to remember its current state, execute the nested instructions, and then restore the original state again. To avoid a tree that grows into infinity without ever stopping, the crayon knows its current
as part of its state. Executing a Call
instruction by default decreases the level by one, and if the level ever would become negative, the crayon instead omits the call.
If you are up to the challenge, from here on you must use the existing profiles as reference to try through trial and error to create your own profile. The content pipeline will complain if there are syntactic errors in your profile.
Collision Detection and Physics
To perform collision detection and physics simulation, it is required to know the general size and shape of the tree. For simple circle-based collision detection,
is a good collision radius to use, as it is the radius at the bottom of the tree.
For more advanced physics, the TreeSkeleton
should provide sufficient information about the tree's topology to build an appropriate representation of the it. There is no official support for breakable branches, but it might
be possible like this. Once your physics library has detected that a branch should break, split the tree skeleton into two parts: the descendants of that branch, and the rest. Then generate new meshes and leaf clouds for the two skeletons. The skeleton generated
from the free-broken branch should then be transformed to the branch's original position on the tree.