In the skeletal animation experiment I created the model by hand, placing one polygon per pixel using an image plane as a template and then manually setting the UVs to match up with the sprite. That was way to tedious and definitely not scalable. I wrote a handy little python function that will take a sprite as input and do all the work for me :-). Check it!
I think I’ll still need to create the “rig” manually. There will be one unique rig per direction, with east/west potentially just being a mirror. After I create one decent quality animation set and I automate the rigging process then I am confident that this will be my pipeline.