3.2.3. Observations

There are hundreds of thousands of fonts in databases, commercial and open-sourced (Google, n.d.; Fonts, n.d.; Inc., n.d.). Those databases seem to be large enough to train vector font generation. Even gathering colossal libraries of font data is apparently not such a problem (Lopes et al. 2019).

Contrary to belief, the dataset size doesn’t require an enormous number of examples to be effective. In fact, previous studies contained as few as 46 font examples (Campbell and Kautz 2014). When developing the font dataset, it is important to set a foundation that can be extended easily.

Figure illustrates the difference between a type designer’s drawing and final font of letter H.
Figure illustrates the difference between a type designer’s drawing and the final font of the letter. The more complex shapes represent even more geometrical details that are not present in the datasets.

A curious observation emerges regarding vector representation: none of the examined datasets utilises original designer drawings from source formats such as .ufo, .sfd or .glyphs. These working files, rather like an artist’s sketchbook, contain valuable geometric relationships absent from the final .otf or .ttf font files. The letter H serves as an illuminating example – type designers rarely conceive it as a single shape, instead constructing it from three distinct components: two vertical stems joined by one horizontal stem. When serifs enter this geometric dance, they manifest as four additional independent shapes. This decomposition into constituent elements reveals crucial insights into the designer’s creative process – insights that vanish like the Cheshire Cat’s grin in the transition to the final font files.

Campbell, Neill D. F., and Jan Kautz. 2014. “Learning a Manifold of Fonts.” ACM Transactions on Graphics 33 (4): 91:1–11. https://doi.org/10.1145/2601097.2601212.
Fonts, Naver. n.d. “Naver Fonts.” Accessed January 13, 2023. https://hangeul.naver.com/font.
Google. n.d. “Google Fonts.” Google Fonts. Accessed January 26, 2023. https://fonts.google.com/noto.
Inc., Fontworks. n.d. “Fontworks.” Accessed January 13, 2023. https://en.fontworks.co.jp/.
Lopes, Raphael Gontijo, David Ha, Douglas Eck, and Jonathon Shlens. 2019. “SVG-VAE: A Learned Representation for Scalable Vector Graphics.” In 2019 IEEE/CVF International Conference on Computer Vision (ICCV), 7929–38. https://doi.org/10.1109/ICCV.2019.00802.

License

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License .

Creative Commons License

Citation

If this work is useful for your research, please cite it as:

@phdthesis{paldia2025generative,
  title={Research and development of generative neural networks for type design},
  author={Paldia, Filip},
  year={2025},
  school={Academy of Fine Arts and Design in Bratislava},
  address={Bratislava, Slovakia},
  type={Doctoral thesis},
  url={https://lttrface.com/doctoral-thesis/},
  note={Department of Visual Communication, Studio Typo}
}