> Also, given a small set of glyphs, would a NN be able to "design" the remainder of the glyphs?
It should be easy enough to train N neural nets, each of which would output one of the glyphs by inputting the rest of them: So a NN for "a", another for "b", etc.
Then a consistent font would be one that outputs itself when fed to the N nets. And a way to create consistent fonts would be to start from anything and iterate until input and outputs converge.
Not exactly what you want, but
"Recurrent Net Dreams Up Fake Chinese Characters in Vector Format with TensorFlow" - http://blog.otoro.net/2015/12/28/recurrent-net-dreams-up-fak...
previous discussion - https://news.ycombinator.com/item?id=10801712
Also, given a small set of glyphs, would a NN be able to "design" the remainder of the glyphs?
At least there is plenty of training data available :)