3.1. Applications: The Use of Generative Models in Type Design

In the ever-evolving landscape of type design, artificial intelligence presents both tantalising possibilities and perplexing uncertainties. The boundaries between human creativity and machine capability remain delightfully unclear, sparking a cascade of questions that dance between practical application and speculative fantasy. Can an AI system truly generate a complete typeface from mere textual descriptions? Might it extract the essence of a logo or illustration to birth an entirely new font? Perhaps most intriguingly, could it serve as a collaborative partner – helping to complete unfinished typefaces or orchestrating novel blends between existing fonts? These questions emerge not from idle curiosity but from a pressing need to understand the practical implications of these tools beyond their theoretical foundations. As the line between human artistry and algorithmic assistance continues to blur, this study endeavours to illuminate what is genuinely possible.

The objective of this investigation is to systematically answer the pressing questions by the knowledge that emerges from the literature database of AI font generation projects. Through comparing the aspirations of type designers with the mastery of machine learning practitioners, shared interests and opportunities begin to crystallise. While – to this day – machine learning models excel at specific tasks rather than serving as universal solutions (Kumar 2022), their focused application proves particularly adept at addressing these fundamental queries about AI’s role in type design.

The results of this investigation reveal a fascinating progression through four distinct applications: blending, completion, and modality translation. This survey traverses from elementary latent space operations (Neill Campbell, n.d.; Carlier et al. 2020) – where fonts gracefully interpolate between styles – to increasingly sophisticated endeavours like style transfer from one font to another (Tian [2016] 2016; Kascenas 2018; Li et al. 2021). The applications demonstrate an evolution in complexity: from basic font mixing to the audacious task of completing partial typefaces (Park [2021] 2022), culminating in cross-modal translations that transform images, text, or even emotions into fonts (Choi and Aizawa 2019; Zhang, Zhao, and Liao 2023). Each application is methodically dissected, with the survey examining not only the technical approaches but also their practical implications and tangible benefits. As one delves deeper into these implementations, it becomes apparent that this seemingly straightforward categorisation serves as a signpost, or the white rabbit at the beginning of the journey.