Machine Learning–Driven Typesetting Inspired by Human Design Practices

Published by Kushtrim Hamzaj
July 10, 2025

The third use case introduces a smart typesetting method that uses machine learning to make intelligent, human-like typographic decisions. Unlike rule-based typesetting, where typesetting decisions rely on hard-coded algorithms, this prototype proposes a system that uses machine learning methods (such as optical character recognition (OCR) and neural networks) to analyze the work of previous type designers and learn from their visual and structural decisions. This approach allows computers to make typographic design decisions based on human logic.

Introduction

To demonstrate the functionality of the smart typesetting system, an easy-to-use interface was designed. The goal of this interface was to prototype the system’s core capabilities in a web environment. On the left side of the interface, there’s a big upload area where designers can drag and drop files, usually scanned images for model training. Above it, there is a icon where user can click and name their model. In the center of the screen, there’s a button called “Train Model.” which triggers the training process. On the right, the output section displays a preview of the generated results. A tab menu lets users preview style settings like Line-Height, Font-Size (which is selected by default), and Font-Width. Below the preview, there’s a gray “Export Model” button for downloading the final trained model.

Main Features of the Prototype

The key feature of the prototype lies in its ability to learn from custom typographic data and autonomously apply those stylistic decisions in real time, directly on the device. It is capable of extracting visual patterns—such as font usage, spacing, and hierarchy—from the provided samples and generalizing them to new content. The following three-step process outlines the complete cycle from training the model to deploying it in a real-world scenario.

Conclusion and Future Work

With the final implementation of the use case, it can be stated that the proposed synergy of variable fonts and machine learning can significantly improve the personalization of digital interfaces in a practical context. While initial feedback on the use case was positive, further research is required to evaluate the effectiveness of variable fonts in adaptive design environments. This includes techniques such as, usability testing, A/B testing, task performance analysis, think-aloud protocols, eye-tracking studies, semi-structured interviews and diary studies to gather insights about each case independently. These tests could further reveal user perceptions and contextual usage insights on how designers perceive each use case.

References:

Rowley, C., & Mittelbach, F. (1991). The Pursuit of Quality How can automated type-
setting achieve the highest standards of craft typography? In EP92 – Proceedings
of Electronic Publishing.

Machine Learning–Driven Typesetting Inspired by Human Design Practices

Introduction

Main Features of the Prototype

Conclusion and Future Work

Artificial Intelligence in the

realms of TYpography & DEsign

Machine Learning–Driven Typesetting Inspired by Human Design Practices

Introduction

Main Features of the Prototype

Conclusion and Future Work

Read also

Dynamic Font Customization Using Real-Time Hand Gesture Recognition with ml5.js and Teachable Machine.

Workshop Recap: Introduction to Artificial Intelligence in the Realms of Design and Arts

Exploring Variable Fonts and Machine Learning for Personalization in Ambient Home Interfaces

Artificial Intelligence in the

realms of TYpography & DEsign