Algorithmic Knitting Design: Integrating algorithmic computing, fashion design and handknitting

Introduction

Algorithmic Knitting Design ¹ is a long-term project initiated in 2020 by composer, interdisciplinary maker, designer and voice artist Stephanie Pan ² integrating fashion design, knitting, and algorithmic computing. Through this project Stephanie develops knitting patterns for apparel and (home) accessories for handknitters, incorporating digital aesthetics into their texture patterns. Alongside Stephanie, the team includes composer/interdisciplinary artist/researcher Stelios Manousakis, PhD ³, composer/software developer Jan Trützschler von Falkenstein, PhD ⁴, and computational architect Weihaw Wang, founder of Sawara Architects. ⁵ There are currently 10 published patterns available. Approximately 3-4 patterns are published per year, with the intention to continue this work and further develop the project.

The design process involves translating algorithms into various knitting techniques such as knit-and-purl textures, fairisle colorwork, intarsia, a new version of ‘marlisle’ ⁶ developed by Stephanie, and lacework, with further research into cable textures. So far, designs have focused on Cellular Automata (CA) as a means for generating texture and colorwork patterns. As Holden and Holden (2021, 443) note, owing to the grid-based nature of this algorithm, “many fiber arts provide natural ways of depicting cellular automata”. ⁷

Algorithmic Knitting Design seeks to: a) present traditional handicrafts, and specifically knitting, as applied science and mathematics; b) take a major step in updating the aesthetics, sensibilities and possibilities of hand-knitting through digital technology; c) democratize and extend the boundaries of ‘high art’ by offering artistic blueprints as accessible knitting patterns to a broad, vast and varied audience typically different from a contemporary art audience. This approach, i.e. providing the instructions for others to create an artwork, is similar to Sol LeWitt’s Wall Drawings. ⁸

Background

Knitting has remained a vital, relevant handicraft and tool through over 3 millennia. Requiring only access to yarn/string and knitting needles, knitting combines portability with practicality, as well as being a pleasurable and soothing manner to clothe and protect oneself from the elements (once adequate proficiency has been achieved). Continuing to help evolve this practice is essential to enduringly recontextualize an age-old craft, with millions of current practitioners worldwide, in current reality. The project centers the knitter in its approach and takes their perspective as a starting point in its priorities.

Knitting patterns can be thought of in similar terms to computer code - more specifically, low-level assembly language - as written instructions that can be read to produce a specified outcome. Textile design and computation are intrinsically connected through a shared genealogy, exemplified by punchcard technology developed by Joseph Marie Jacquard in 1803 to control the Jacquard Loom. Punchcards were adopted by Charles Babbage in 1837 as a mechanism for storing data and math operations for his Analytical Engine and used subsequently by Ada Lovelace in 1843 to create the first computer program - a methodology that remained fundamental to computing until the 1960s. ⁹ While this connection may be clear within scientific and research communities, it is generally forgotten or even unknown to general public and the crafting community at large. Algorithmic Knitting Design aims to revisit and revive this link between computing and fiber arts, examining how computational advances - and in particular algorithmic thinking - can help us redefine the traditional art of knitting within a contemporary aesthetic.

Related work

Recently, there has been a growing interest by scientists and researchers in exploring how knitting can inform scientific research. Janelle Shane’s experiments with Machine Learning ¹⁰ and the research of mathematician and physicist Dr. Elisabetta Matsumoto and her colleagues at Georgia Tech ^{11 12 13} are examples of research through knitting design, investigating how old applied knowledge can inform and evolve new knowledge and theoretical thinking.

Within the realm of fashion and wearable design, a number of fashion designers are integrating new technologies into their work, most notably Iris van Herpen. ¹⁴ Knitwear is an inherent part of the commercial fashion design sector, and there are many designers using digital design techniques for knitwear within their collections. Independent specialists in knitwear include Fabienne Serriere and her CA design work as KnitYak ¹⁵, and Sebastian Kox aka monobrau’s glitchwear. ¹⁶ Tech-driven experiments in knitwear are almost exclusively limited to machine knitting that allow for high levels of intricacy, complexity, automation, and rapid fabrication. In practically all such cases, the product is fully fabricated and the patterns are not available. This creates a closed proprietary loop concealing the blueprint of the design - an understandable approach given both the commercial nature of the industry and the complexity involved in computational design.

In textile art applications, it is fairly common practice for large scale textile works to include community contributions and shared creation by and with crafting communities, and are often part of a communal feminist practice. This is also true in more tech-driven instances, e.g. Margaret Wertheim’s Coral Reef Project ¹⁷, Pei Ying Lin’s Studies of Interbeing – Trance 1.1 – Knit a Spike Protein ¹⁸. The resulting work of the crafters ultimately belong to the artwork, with patterns shared with individuals being parts of a whole, rather than a standalone piece.

Access to available contemporary, digital design patterns for self-fabrication for the handcrafter is severely lacking, with limited offerings to computational, digital approaches. One such example is atacac’s Sharewear platform which contains downloadable sewing patterns generated with 3D modelling and similar digital tools. ^{19 20}. In the last 25 years, a handful of knitwear designers and researchers have been investigating the utilization of CA to create publicly available handknitting patterns. Of note are Debbie New’s designs, likely the first published applications of CA in knitting ^{21 22}, a shawl pattern by well-established knitwear designer Norah Gaughan ²³, and designs by Lana Holden based on a special type of CA she developed for knitting. ^{24 25} However, these applications remain limited in scope.

Philosophy and motivation

In contrast to the closed loop of fashion, Algorithmic Knitting Design develops knittable patterns with high fashion aesthetics that are publicly accessible to all knitters. Designs are inspired by the Belgian and Japanese minimalist approach to form and construction, and by digital aesthetics for texture and colorwork patterns. The project opens up the field of handknitting through technology and offers a new proposal for evolving it, allowing knitters to become active participants in executing / fabricating algorithmic designs with a contemporary aesthetic, without requiring them to know how to code. The project taps into a field of traditional arts with built-in crossover potential to algorithmic thinking and skilled practitioners who have the ability to understand, interpret and execute complex forms and patterns.

Approaching this project with the feminist concept ‘the personal is political’ ²⁶, the need to bring these two worlds together, algorithms and knitting, comes from two major motivations:
* The desire to challenge notions and understandings of what is ‘scientific’ or ‘technological’ versus what is not, and specifically to posit knitting as a form of applied science from the perspective of the knitter, as opposed to an academic approach. * A personal desire to wear and create complex and challenging knit apparel that is contemporary and takes into account digital aesthetics.

Algorithmic Knitting Design approaches the integration of technology and handicraft as research for design: a craft-centric examination of what science can mean to traditional craft. It aims to contribute to the two-way discourse between algorithmic thinking and knitting from this antithetical point of view that emerges from the perspective of the knitter, rather than of the scientist.

Workflow

Stephanie and Stelios initially established a workflow for designing knitwear and developing knitting patterns based on open source software for algorithmic design, in conjunction with knitwear designing software Stitch Fiddle. ²⁷ Proof-of-concept tests involved Mirek Wojtowicz’s Cellular Automata simulator Mirek’s Celebration, or MCell. ²⁸ This was soon replaced by Stelios’ development of a Cellular Automata parametric and algorithmic design environment in the SuperCollider programming language ²⁹, using Yota Morimoto’s port of Wojtowicz’s CA code. ^{30 31} The process has further been streamlined by Jan’s conversion of Stelios’s implementation into a standalone app, Knitting Factory. ³² This enables Stephanie to design autonomously using this app together with Stitch Fiddle.

The knitting design process currently involves the following steps.

1. Shaping, construction, and knitting technique(s):

• Developing the overall shape and construction of the garment or accessory to be knit, as well as the knitting techniques to be used.
• Choosing the preferred yarn type and needle size to determine knitting gauge. ³³
• After establishing the gauge, calculating the desired ‘world size’ for the cellular automata software; this corresponds to the number of rows and stitches required to achieve the desired measurements of the item.

2. Populating the surface algorithmically:

• Running the software to create a selection of interesting and aesthetically appealing iterations of various rules. This operation involves experimenting with and evaluating a variety of transformation rules, which establish the general form of the fractal pattern that will be generated. It is currently possible to iterate patterns using a single rule, to dictate specific moments where the software changes rules, or to instruct the software to change rules algorithmically using parametric design principles. This opens up many different possibilities such as stipulating how many rule changes may occur within a pattern, which rules to change to, and how long each rule should be iterated for. All these parameters can be defined, or set algorithmically.
• Once a rule or subset of rules is selected, experimenting with different seeds (i.e. the starting state which these rules iteratively transform). Seeding principally affects the density of the texture, and the placement of the cellular automata pattern on the knitted canvas.
• Thereafter an image file (.png) with precisely the same pixel count as the world size is exported and imported into Stitch Fiddle.

3. Translation into a knitting chart:

• Within Stitch Fiddle, the image file is imported as a black and white image file with the exact same pixel count (where cells equal stitches).
• Thereafter the resulting chart is manipulated to transform the image to fairisle colorwork (simple 1-to-1 change of black and white cells to appropriate colors), knit-and-purl (i.e. black cells become knit stitches and white cells become purl stitches, with the exception of vertical lines, due to the nature of purl stitches), intarsia + knit-and-purl (requiring more manual intervention, as this requires at least 4 different types of knitting commands vs. the 2 commands present in the image), marlisle (requiring 3 types of knitting commands and artistic interpretation of the image file). Other repeating textures may replace given sections to add interest or variety (i.e. simple lace and honeycomb stitch as in Algorithmic Diva Bolero and Event Horizon circle scarf). In certain cases, multiple CA output files are combined to create the eventual design; these may be adapted and modified based on aesthetics and other knitting considerations (Figure 7). This approach, not being dogmatic or attached to ‘algorithmic purity’, is comparable to algorithmic composing pioneer Iannis Xenakis’s documented practice of modifying the output of his algorithms based on aesthetic considerations. ³⁴

4. Hereafter, the general development process for publishing a knitting pattern for general public is followed:

• Prototyping: designer knits a sample of the pattern, making adjustments where necessary (1-2 months).
• Blueprinting: If satisfied with this sample, she then writes the knitting pattern (written instructions, schematics, knitting charts, sizing, garment assembly, modification options, etc.) (1-6 weeks)
• Testknitting: a group of testknitters is assembled to knit the pattern (2-4 months). ³⁵
• Publishing: photos of the sample are taken, the pattern is published on the Algorithmic Knitting Design website and Ravelry page. ³⁶

Some reflections on translation

The binary nature of the 1-D cellular automata algorithms we have been focusing on so far have some limitations: Their output lends itself very well to knitting techniques that are similarly binary in nature— knit-and-purl textures, fairisle colorwork, intarsia and to a certain extent, marlisle. Different CA rules lend themselves better to different techniques. Iterations with large swaths of 0’s translate well to knit-and-purl textures, intarsia, and marlisle, but are problematic for fairisle, requiring long chains of ‘catching floats’. ³⁷ Vertical lines must be adjusted to appear correctly in knit-and-purl textures. ³⁸ Highly chaotic textures generally translate best in fairisle colorwork.

However, when translating to non-binary knitting techniques, such as lacework and cable textures, much more creative, manual and labor-intensive interpretation of the CA output is necessary. CA lacework requires manually translating the image file in Stitch Fiddle cell-by-cell and stitch-by-stitch requiring at a minimum 3 types of stitches (of which 2 are interdependent), and more realistically at least 7 different types of stitches (5 of which are interdependent). Charts run between approximately 60-200 stitches per row times 100-400 rows in total, or upwards of 80,000 stitches, depending on the item (lower end is for smaller cowls, higher end is for largest sizes of garments). The other challenge with lacework is that it is fairly unpredictable and the charts are not very indicative of the knitted result, and thus require much more prototyping (i.e. handknitting samples/swatches) which is labor-intensive and time consuming.

For all knitting techniques, there is the overarching challenge of balancing complexity with executability and enjoyment. Finding ways to incorporate sections of ‘mindless knitting’, i.e. repeating patterns such as stockinette, moss stitch, rib stitch, etc. into the patterns are generally good ways to increase executability and balance required focus and practical portability, as charts are generally very large, and generally most readable when printed.

Future work

Current research with Stelios includes exploring L-systems for generating lacework, but still ultimately runs into similar challenges as cellular automata. The largest challenges are: developing an algorithm that can account for the interdependency of lace stitches, and still output something visually aesthetic; the time- and labor- intensive nature of prototyping in handknitting; outputting a readable chart for publication – Stitch Fiddle is only equipped to translate images into colorwork knitting charts, not interpret symbols from imported files. While this is easy enough to manage when translating the chart for knit-and-purl textures as well as intarsia, the workflow breaks down for more complex techniques such as lacework and cables. Research into cable textures is being done with Weihaw, developing a visualization software for designing with cables. Similar to lacework, cable knitting charts offer little insight into the actual end result of the knitted fabric. Weihaw’s software will allow for visual placement of cables, after which point it will output a knittable chart with the appropriate cable stitches and world sizes. Further future plans include a potential residency at the Making With… cluster of the Department of Industrial Design at TU Eindhoven ³⁹, through Dr. Kristina Andersen, where they have a textile lab, including a Kniterate digital knitting machine, which would make prototyping and experimenting with techniques much more efficient.

Conclusion

The project ultimately aims to take a major step in updating current expectations and aesthetics in the knitting community by challenging knitters’ understanding of their own craft. It balances contemporary high fashion aesthetics with practicality and wearability, and large, non-repeating patterns with executability. Algorithmic Knitting Design asks an open-ended question: if we start to understand ourselves, knitters, as scientists, as extraordinary physical computers, what happens to our approach to the craft, our ideas on what we can do and accomplish, and what new challenges can we take on?