neural network gerstner
we can encode any series of typographic combinations as a list of
dFontFamily, dFontWeight, bFontFamily, bFontWeight, bFontSize, dSizeMultiplier, dMargin, dIndent
[0, 0, 0, 0, 0, 0, 0, 0]
how about processing a seminal piece of typographic design like Schiff Nach Europa into paragraph-level training data - their own typographic choices, and the relationship between paragraphs?
This all seems serializable; if it is we can train a neural network to design in this style. relationships between text length and font size, or line length etc.