In addition to storing real data like a mesh or a curve, objects can store instances, which themselves can reference more geometry, an object, or a collection. The purpose of instancing is to allow duplicating geometry and storing it in an object, without duplicating the actual data. This optimization allows render engines like Cycles to handle the same geometry data in many different locations better than when the data is duplicated.
Each instance keeps track of which geometry it corresponds to
and how the instanced is transformed compared to it's source geometry.
Instances can also store the
id attribute, which is used to correct motion blur when instances move in an animation.
Instances can be created with geometry nodes using the 实例化于点上.
当前，从几何节点实例化不能与从属性编辑器中的 实例化 面板实例化混合。
Since instances can store a geometry, and a geometry can contain instances, nested instancing is possible. In other words, it is possible to instance an instance, or even a collection of instances. For example, by default, the 实例化于点上 will create nested instances by instancing instances on the points of real geometry.
Here, nested instancing is used to distribute geometry that contains both a mesh and instances. The output geometry contains a "real" mesh and a group of instances. Each instance contains a sphere mesh and many instances of a cone geometry.
What makes this method helpful is the output geometry only contains three unique meshes: the plane, the sphere, and the cone. This would make the performance much better if the meshes were more complicated.
Only eight levels of nested instancing are supported for rendering and viewing in the viewport. Though deeper trees of instances can be made inside geometry nodes, they must be realized at the end of the node tree.
The term "realizing" instances referes to converting the instances into unique geometry. When instances are realized they will take up more memory and manipulation to geometry will have to be processed individually rather the once per instancing geometry.
To realize instance use the 实现实例.
Almost all nodes that process geometry do so by processing each unique geometry separately rather than realized geometry. For example, if a 表面细分 was placed at the end of the example above, it would only have to subdivide three meshes, rather than each instance of a mesh. Another important example is processing with the output of the 字符串到曲线, where each unique character only has to be processed once.
这种方法可以大大提高性能，但是这意味着操作的结果对于特定几何体的每个实例都是相同的。为了每个实例都有独特的结果，可以使用 实现实例 节点。