To design character Models for optimal rendering and animation speed:
Using these techniques might help increase the animation and rendering speed of your characters, but be aware that they might also reduce the visual fidelity and realistic effects you are trying to achieve. There is no single answer for every situation that can produce a perfect balance between better performance and visual realism. You have to find the perfect balance according to the complexity of your characters and your scene, vs. the overall appearance and realism that you require.
Use only a single skinned Mesh Renderer for each character. Unity optimizes animation using visibility culling and bounding volume updates, and only activates these optimizations if you use one Animation component and one skinned Mesh Renderer in conjunction. Using two skinned Meshes in place of a single Mesh could roughly double rendering time for a Model, and there is seldom any practical advantage in using multiple Meshes.
Keep the number of Materials on each Mesh as low as possible. Only have more than one Material on a character when you need to use different Shaders for different parts, such as a special Shader for the eyes. However, two or three Materials per character should be sufficient in almost all cases.
In general, the fewer bones you use, the better the performance is. However, sometimes you need to create characters with a large number of bones: for example, when you want a lot of customizable attachments. These extra bones increase the size of the build, and may have a relative processing cost for each additional bone. For example, 15 additional bones on a rig that already has 30 bones takes Unity 50% more operations to solve in Generic mode, which might impact the time it takes to solve.
Note that you can have additional bones for Generic and Humanoid types. When you have no animations playing using the additional bones, the processing cost should be negligible. This cost is even lower if their attachments are non-existent or hidden.
For performance reasons, use linear blend skinning with a maximum of four influences per vertex. However, some 3D modeling applications allow more than four bones to influence vertices, so you have to weigh the performance cost against greater control. For more information, see Importing skinned Meshes.
The number of polygons you should use depends on the visual quality you require and the platform you are targeting. These two competing facts are equally true:
Also consider what else is competing for rendering resources. If you have a lot of GameObjects or Terrain objects on screen at the same time, consider reducing the polygon count per Mesh.
Desktop platforms can generally handle more complexity than mobile platforms, but performance on mobile platforms can vary enormously depending on the quality and capacity of the device itself.
When Unity imports animations, it bakes a Model’s inverse kinematic (IK) nodes into forward kinematics (FK), and so Unity doesn’t need the IK nodes at all. However, if they are left in the Model, then Unity still includes them in calculations, even though they don’t affect the animation. You can either delete the redundant IK nodes in Unity or in your 3D modeling application. To facilitate removing the IK nodes, keep separate IK and FK hierarchies while modeling.