What are the advantages to use per vertex motion (Geometry nodes), transform motion(Transform nodes) instead of just averaging final frames?
Imaging there are a moving camera or/and moving objects in the scene, by averaging rendered images in a time interval, we could generate motion blur effects automatically. Sure, OptixMotionBlur shows a use case, but I still do not quite understand why they are useful and when I should use that.
Maybe I missed something important,
When rendering final images at discrete time steps and then average them, you’ll get visible banding like on the right side of this image: http://images.slideplayer.com/15/4786242/slides/slide_10.jpg
The “cook” example inside the older OptiX SDK examples did that for motion blur and depth of field.
You would have to render many final images to fill the motion this way, which is only feasible if your rendering is really fast.
In contrast, a stochastic motion blur implementation with a time per ray is able to sample the time continuously in a progressive algorithm which produces smoother but initially noisy results like any Monte Carlo algorithm.
Additionally it allows for camera shutter effects like these in a single launch: