Image generation deals with obtaining the individual frames of the animation. This usually involves adding extra features to an existing visualisation to make it dynamic and to save these frames to disk in some format.
Where the data being visualised has a temporal dimension, or where time is being used as a visualisation dimension, the most common and simplest way of generating images is to simply time step through the data. The time step delta used is usually the sampled rate of the data, but may be larger than this if interpolation is acceptable6.
The main disadvantage with this method is the frame rate is fixed at the rate the images are generated at. When the number of generated images is low, this can result in an animation that isn't smooth, however generating many frames can be time-consuming (and the problem remains that the number of frames is fixed).
The more sophisticated solution to this problem is to use a method of animating called key-frame animation. This is similar to the above method, except that only certain ``key'' (important) frames are chosen and the system's state at these frames stored. Frames required between these key-frames are interpolated from the system's state at the key-frames, usually linearly7. This has the advantage that the animation can be created from the key-frames at the frame rate of the output device. However, the disadvantage is that key-frames and the interpolation method should be chosen carefully, especially in a scientific visualisation setting, to avoid inappropriate or misleading animations from being generated.
In turn, key-frame animation leads on to the still more sophisticated areas of kinematics and dynamics (and inverse kinematics and dynamics). These are methods of simulating a system according to physical laws to provide a still more accurate and visually appealing animation.
Where frames are generated from the visualisation (in any of the above techniques), the problem of temporal aliasing occasionally comes up. This is where phenomena in the visualisation are occuring on a time-scale smaller than the time step delta of the animation. The result is that incorrect or misleading animations may be generated, with the common example being the way human eyes may perceive the wheels of a travelling car to be spinning backwards when they are actually spinning forwards at the correct frequency.