This script is a nice idea - but its completely useless. The morph targets simply are wrong from the view of facial expression research. This is why you will get useless expressions with this approach. In facial expression research single muscle movements are used - this is is only possible when all the appearance changes caused by muscle movements are modeled as separate morph targets.  Usually this is done by using the facial action coding system which explains in detail the appearance changes caused by a muscle contraction. Another problem is that most of the expressions are asymmetric - so you will need morph targets for each side of the face.

The next step then is to figure out which muscles are moved when generating an emotional expression. There are theoretically two ways - one is to rely on the concept of emotional expression families (six basic emotions) and use the dials of the muscle morph targets to generate the combinations. These basic emotions can be mixed. Another one is to uses muscle activation data from a three dimensional emotion space where every muscle has an activation in terms of dominance, pleasure and arousal. This is how it is done in science, many computer games and professional animation.

Now comes the tricky part - there are defined and published rules which muscles can not be contracted together at the same time - so you need a script for checking. This script is not trivial.

The gif animation contains 100 random expressions generated this way. The model V2 has 30 facial muscles (no asymmetry), but we have this for V3. The morph targets have been done in 3dsmax and cross checked for validity several times.

If somebody is not interested in learning all these things we have a python script which uses a genetic algorithm to produce expressions or head forms or whatever you want. We will make this available for free when the final testing is through.