Does training image generators on specimens of typefaces, with associated captions like name and style of the typefaces, improve their generation of text in images, without needing font files? (It should be noted that font files are required for exact reproductions, but it doesn't need to be exact).

Full article content could not be extracted automatically. Read the original below.