Research project, ECAL 2021


AIZI / AI字 / 爱字: Artificial Intelligence-Aided Type Design

A Latin typeface contains a mere 400 to 700 glyphs, on average. Thanks to the alphabetic system, these relatively few characters can spell out every word that exists in a given language. A single style of a Chinese font must contain a minimum of 6,763 commonly used glyphs to be of any use, while typefaces for literary use, for instance, need thousands more. Famous Chinese historical dictionaries contains between 47,000 and 54,000 hanzi (characters), the most complete going to the incredible number of 106,230.

Thus, creating a Chinese can only be achieved though long time and team effort. Those two elements are inhibiting creativity. First, it makes type foundries tend to not take risks with their design. Second, it prevent freelance designers (Chinese or not) to design for the script, which volume makes it unattainable. This situation cuts type designers for a script used by 900 million native speakers of Mandarin language, and around 400 million second-language readers.

AIZI Research projects is a collaboration between ECAL and EPFL Computer Vision Laboratory. Its aim is to develop an artificial intelligence tool to help the creation of hanzi. The idea is to train an AI to generate glyphs from a small number of ‘seed’ characters, using Generative Adversarial Network (GAN): two algorithms fighting each other, endlessly attempting to outperform one another. Chinese script is rich of thousands of hanzi, but their construction is, on many point, very logical and systematic. It process by assembling a limited number of radicals in order to produce new signs. For the designer, a major difficulty is that depending of the surrounding components, the design of the radicals changes, always self-adapting to the context, in order to achieve harmonious forms. 

The output of the research project is a database, of the different glyph constructions that exists in Chinese. It currently covers 90,000 hanzi, and will be freely accessible online, as a encouragement for the type design technological scene to develop. The project was also the occasion to think about the role of computer-aided design in the field of typography, its future development, and its application within the field of type design education.

Main applicant
ECAL/University of Art and Design Lausanne
Matthieu Cortat
(chef de projet)

Research team
Kai Bernau, Mathieu Salzmann, Shuhui Shi, Wei Wang

January 2020 – June 2021

Supported by 
ECAL/University of Art and Design Lausanne
Strategic fund of the University of Applied Sciences Western Switzerland (HES-SO RCDAV)

First presentation of AIZI project

@ ECAL / Calypso Mahieu

Average cases. The algorithm shows better results in lower resolution images and could be transfer to larger size via affine transformation

© ECAL / Shuhui Shi

AIZI Database

© ECAL / Shuhui Shi

AIZI Algorithm model

© ECAL / Shuhui Shi