Crystal type: Physiomorphic font design through generative tooling

On view from May 3-7, 2023 at OCAD GradEx 108

My project explores the role of design tools to creative outcomes through applications of data science in font design, and challenges the way we communicate through digital typefaces.

It begins with an organic phenomenon and develops through a series of experimentations with man-made tools. I was inspired by the formation of salt crystals to grow my own using physical letters. These letters were then brought into the digital realm, where I played with creating my own font, using a machine learning model to produce characters given reference input glyphs, and displaying the font on a website using OpenAI’s speech to text Whisper API.

Natural and artificial crystal type


I describe the concept of my font as “physiomorphic”, a term discussed by Graham Richards in his 1989 book On psychological language and the physiomorphic basis of human nature, where he proposes the theory that the terms of psychological language, that is, words for internal states, are derived from physical terms based on our environment and surroundings. For example, the word “stonyhearted” is based upon the metaphorical use of the two physical nouns “stone” and “heart.” Etymologically, “physiomorphism” is derived from the Greek words “physio” (meaning “physical” or “natural”) and “morphe” (meaning “form” or “shape”). In Crystal type, I make digital fonts which bear physical characteristics – carrying a duality of natural elements and machine-aided qualities. 

As early civilizations were built and sustained from the natural environment, much of their written language was a manifestation of physical elements. The Sumerian civilization, lasting from 4100 to 1750 B.C., used pictographs to represent concepts such as “ox”, “sun”, and “grain”. Local materials of clay and reeds were used as tablets and styluses, becoming some of the first tools in producing writing systems.

Evolution of Sumerian writing system. Source: The Oriental Institute, the University of Chicago, via Encyclopedia Brittanica

The role of tools in capitalistic systems

I believe that humans are intrinsically tool makers. Languages, writing systems, and fonts are all tools for communication that have fundamentally changed the world. And, as we change the world, the world changes us. Now more than ever, our lives run on battery life, computational power, and storage space. We send texts filled with acronyms we invented, scroll social media feeds curated by algorithms, and spill our thoughts to chatbots trained on parameters of data. My work questions the role of technology in the things we create.

Some of the biggest technological changes in history have been driven by money. The Industrial Revolution of the late 1700s saw major transformations in manufacturing and automation, allowing companies to increase profit margins, albeit often at the expense of blue collar workers. Shifting from individual craftsmanship to factory-based mass production enabled goods to be made more quickly, but lead to greater specialization of labour. As people developed more focused skillsets to accumulate capital, they became less self-sufficient, relying on others for the goods and services they needed. The once individual practice of type design became a mechanically driven process in which designing and making were separated, where different teams were responsible for design, punch cutting, type setting, and more. It wasn’t until the introduction of personal computers in the late 20th century that digital type design became necessary, which Matthew Carter points out, reunited designing and making, providing the ability for the independent designer to also produce and distribute their own fonts. 

To me, Crystal type is both independence and relying on other people. I can say I made physical letters, I made tools, I made digital fonts myself. But I would be disregarding the type designers who laid the foundation for the characteristics of the Latin alphabet, the forces behind the intricate workings of nature, the creators of the programming languages I write in, the researchers who contributed to the machine learning models I use, the sources of data they’ve been trained on. To be able to create independently is the result of the technology that precedes me and the people who advanced it.

Creativity in the age of generative technology

Over the years, the term “generative” has taken on different meanings across a range of disciplines. In math, generative functions typically produce sequences of numbers or other structures based on a set of rules. For example, the Fibonacci sequence is generated by a recursive function, in which each of its numbers are the sum of the two preceding numbers: 0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, and so on. In linguistics, Noam Chomsky introduced the concept of generative grammar in his 1957 work Syntactic Structures. He proposed that language is a set of recursive rules that can be used to generate an infinite number of grammatical sentences. And today, the use of generative machine learning models in creative tools has seen explosive growth.

At its core, generativity is about creation via rule-based systems. We began with writing binary instructions, which have turned into more advanced algorithms, and now we develop machine learning models that can create their own rules by finding patterns in large amounts of data.

The use of generative models in font design intrigued me. After going down a rabbit hole of academic papers and code repositories, I saw the potential in applications of machine learning in font design, such as few-shot glyph synthesis and multilingual style transfer. But I was also wary about the results of these approaches. Could I make a font of the future, with tools trained on data of the past? Much of the output from today’s trending technology feels like a recycled version of yesterday’s creations. New tools give us new capabilities, but also present new limits. Would I be limiting myself to making remixes of what already exists?

I was thinking about this question when I came across a 2017 conversation with Laurel Schwulst on The Creative Independent, where Charles Broskoski, a co-founder of, shared, “I recently learned Einstein used to do something specific. To get his brain in the zone, he would bring together two very different things to see what the connection was between them.” This is not far from the sentiments of Steve Jobs expressed in I, Steve: Steve Jobs in His Own Words, a collection of quotes edited by George Beahm, where he says, 

“Creativity is just connecting things. When you ask creative people how they did something, they feel a little guilty because they didn’t really do it, they just saw something. It seemed obvious to them after a while. That’s because they were able to connect experiences they’ve had and synthesize new things. And the reason they were able to do that was that they’ve had more experiences or they have thought more about their experiences than other people.”

In a commencement speech to Stanford’s graduating class of 2005, Jobs recalls deciding to take a calligraphy class in college after seeing posters and labels around campus. He described it as both fascinating, yet without “even a hope of any practical application in my life.” But, ten years later, this experience inspired the typography of the Apple Macintosh, which became one of its defining features, and set it apart from other computers of the time.

In the pursuit of making something entirely new, I realized that most of the ideas I considered “creative” were actually the result of seeing old things in new ways. Part of what makes humans special is our ability to make connections based on our lived experiences, and how we express them through the tools we choose. Every detail of our creative practice reflects a part of ourselves and makes a difference on our art. So I thought, to create a new font, I should try a new process with new tools.


I centered my next experiments around two philosophies:

  1. Find unexpected connections

  2. Use unconventional tools

An early step in this direction was a font I made in Figma, a tool primarily used for designing interfaces. Standard software for font design is usually platform-specific, built for individual use, and costs over a few hundred dollars. Figma, on the other hand, is browser-based, collaborative, and free. I wanted to introduce an alternative tool to encourage different approaches to font design and lower barriers to entry.

Then came my project, Text Decorator, which combines typography with prompt-based image generation. Using Hugging Face Spaces, Python, and Gradio, I built a prototype for “decorating” text input, powered by Stable Diffusion’s Image-to-Image pipeline. It receives a text, font size, and prompt; an image is created from the text and font size, which is sent to the model alongside the prompt, producing an output that is a mixture of both the initial letterforms and prompt image. 

While pixel-based image generation was taking off in the mainstream, comparative technology for vector graphics and fonts were not as widely available, and relatively less researched. To understand what current models were capable of, I attempted to run a few of them on my computer. I had the most success with the 2021 implementation of DeepVecFont by Yizhi Wang and Zhouhui Lian, which is a few-shot model for synthesizing an entire alphabet using a small number of reference glyphs.

Output of DeepVecFont with input glyphs A, B, and I in the top row, and synthesized glyphs underneath

Since the model’s output is a byproduct of its training on existing fonts, I looked for design inspiration outside of fonts. It was winter in Toronto, and my mind wandered to the shapes of ice crystals and the way they refract light. The idea of combining a natural process with a machine-aided practice felt like a connection that was unexpected, yet made complete sense to me. I found a tutorial about growing salt crystals with borax powder and started my science project, experimenting with materials and techniques to yield different results.

After landing on a set of physical letters to serve as the source of my digital work, I then developed a collection of alphabets:

I further refined the Traced font by including stylistic alternates for the ‘E,’ ‘F’, and ‘H’ characters, adding punctuation, and extending it into a variable font, which allows for an unlimited range of possibilities between the values of 100 and 400.

Living in the age of data has also opened up new forms of communication that go beyond static text. Just by capturing a voice, we can access a mountain of information such as transcriptions, sentiment, volume, frequency, and more. In exploring these capabilities, I made a website that displays text from audio input using OpenAI’s speech to text Whisper API, rendered in the Traced font.

Interactive web demo

In a sense, people on the website are simultaneously talking to: 

  • Themselves: reading what they have said 

  • A computer: facing a screen that contains a (not always accurate) model’s interpretation of what they have said 

  • Me: through their interaction with the font I designed 

Through the combination of natural growth and machine learning, my project questions what it means to make and break the rules of generativity in the process of human-led creation. Crystal type experiments with the use of data driven technology to make connections across disparate concepts, and expresses the constantly changing state of how the world communicates.


Although Crystal type is, to my knowledge, the first project which explores the use of generative and language models in conjunction with a font based on the formation of crystals, there are a number of related works which inspired me: 

  • Occlusion Grotesque (2021), a typeface by Bjørn Karmann created through carving letters into a tree and using the data collected to train a model on how a tree grows and deforms letters over time 

  • Laika (2009), a variable font by Michael Flückiger and Nicolas Kunz which changes axes through real-time data such as stock exchange prices, the weather, and information from physical sensors

  • Phase (2019) by Elias Hanzer, a component-based generative type concept living on a website, which serves as a tool for others to customize the font through audio input and digital interaction

  • Computers and Creativity (2021) by Molly Mielke, an essay about harnessing digital creative tools in human innovation

Thank yous

This project would be very different without the help and advice I received from:

  • My thesis advisors Adam, Immony, and David

  • My Digital Futures thesis classmates

  • Yihui, Claudia, Kelly, Yizhi, James, Arsh, Bria, Henry, Daniel, Stella

Posted 04/25/2023

Last updated 05/03/2023


in Toronto

Listening to