Well, I'm getting a better grip on the elements which are important in belt simulation. However I don't seem to be getting consistent results, but that may be because my expectation of consistency is wrong. :)
Anyway, here's a quick summary of the points. Some of these are repeated from the advice previously given by aRtBee and others. Apologies if it seems that I'm trying to take credit for that advice.
- Make the belt used in the simulation follow the figure's surface as consistently as possible. This is because the cloth will be squashed between belt and figure as the simulation proceeds, and the simulator doesn't like it if that's overdone. I used Cage's Mr. Skinner script to make the belt, with an offset of 0.001 or 0.0025 (see later discussion). Bear in mind that you can change the sim-belt for something more detailed afterwards.
- Have lots of triangular polys in the cloth. I triangulated my original quad mesh using UVMapper tools > facets > triangulate (at polygon centre) to give 11952 polys.
- In simulation settings, check both 'Object vertex against cloth polygon'
and 'Object polygon against cloth polygon' options.
- For both figure and belt: set collision offset to minimum (0.1)
and collision depth to 0.001.
- Increase the steps per frame as required - for this picture I went to 16 (giving ~16s per frame for 30 frames). However I wasn't wholly scientific about it, and changed more than one parameter at once. I need to try reducing this figure again to see what happens.
The render attached was made after a simulation with the belt offset being 0.001. You can see Antonia poking through the cloth a little, presumably as a result of having such a close approach of the belt to the body.
I then increased the belt offset to 0.0025, on the assumption that it ought really to be twice the collision depth. However the cloth then started poking through the belt, for some reason. As I said, inconsistent. I have to put this aside for now, but I'll carry on when I'm able.
