weaving
fragments
into threads

I: MODEL-WIP: The Experiment

In February 2024, we set out to find an artist who was curious to experiment with AI in their practice. In our search, we found many kinds of artists.

​Painters, writers, illustrators, photographers, knitters, weavers, storytellers, filmographers, worldbuilders, sculptors and performers.

​Several summarized their viewpoint on AI with the same two emotions. They were both terrified and excited.

​Terrified that AI would steal or compromise their visual identity.

​Excited that this tool would open the door to new and unexplored possibilities in their practice.

​So we identified an opportunity: To put AI tooling into the hands of creators, enabling them to confront their fear through experimentation. To close the gap between how models currently perform and how artists would like for them to perform.

​To take these AI systems that seem so enormous and make them more bespoke.

MODEL-WIP = A community experiment to fund the creation of artist-owned AI models.

​MODEL-WIP is an artist commission group backed by SuperRare, Base, TITLES, and Zora, funding and partnering with artists to explore their relationship, ownership, and practice with AI.

With the rise of generative AI and Ethereum as rails for attribution and ownership of creativity on the internet, MODEL-WIP is an experiment at understanding the intersection of these two emergent technologies and how they can be used to benefit the artists and creators using them.

We started with a simple question: How can artists own, define, and monetize their likeness in the age of AI?

It turns out that artists are not very interested in talking about their “likeness”. Their likeness, or style, often varies significantly from project to project.

Instead, they wanted to talk about their identity.

And about cultural preservation.

Senegalese flowers, Cameroonian fashion, Mongolian languages and Spanish literature.

These are just some of the ideas that came up in conversations with artists about training an AI model, in addition to questions like:

​How can an AI model help to preserve culture ?

How can an AI model help to tell an artist’s truth?

Is there art to be found in a model’s failure to replicate a certain artistic style?

II: Emily Xie

Emily Xie’s willingness to experiment and push the boundaries of AI as a tool for art-making made her the top choice for MODEL-WIP.

A digitally native artist living in NYC, Emily works with code and computation to create digital textures and forms inspired by textiles, collage, and wallpaper. Her pieces explore the interplay of materials and patterns, weaving together themes of mythology, memory, tradition, and heritage.

Emily’s generative systems often balance many contrasts such as randomness and control, tradition and modernity, and tend to incorporate skeuomorphic techniques to reimagine materiality. Xie often contemplates the intersection of computation and textile arts, drawing on the historical and sensory richness of fabrics and mixed media.

Emily’s creative coding work is collected and shown internationally. Recently, she has exhibited at the Untitled Art Fair, the United Nations Headquarters, Singapore ArtScience Museum, Kunsthalle Zürich, Unit London, and the Armory Show.Prior to pursuing art full-time, Emily built a career as a software engineer, and slowly explored creative coding on the side.

Language emerged as a central focus for Emily throughout this experiment: both her proficiency in programming language, as a tool for self-expression, and her quest to further comprehend written Chinese language: an important piece to her multi-cultural identity as a first-generation immigrant who largely grew up in the US.

The Process

Phase I: Breaking Apart

Emily focused the early experimentation phase of this project on Seal Script characters, given their pictorial nature. She was particularly interested in the graceful, elegant lines that composed the characters, and how they occupied an in-between place that triangulated abstraction, representation, and meaning. She started by writing a program that compiled a vast selection of SVG characters from the public domain.

From there, Emily mapped the characters onto the viewport in her program using a recursive method of partitioning the canvas. She liked how this structure subtly recalled the flow of a traditional scroll. She then built optimizations around the path transformations, allowing her to easily experiment and prototype with minimal wait time.

Once the initial scaffolding and engineering work was complete, Emily set out to work on the more stylistic components. Her first impulse was to destroy the characters entirely – particularly given her personal relationship to them. These shapes, while foundational to the Chinese characters that held such deep meaning to many, were semantically meaningless to Emily. Destroying them offered a way to grapple with this disconnect, and allowed her to reclaim them as forms of her own.

She started by first deforming the characters: distorting and warping the shapes – a sort of “tug of war with the computer”: taking clean, systematic and machine-like edges and trying to bring in a messy, organic nature to them.

As she continued to experiment and explore methods of mathematical fragmentation, Emily wondered what might happen if these shapes were to be analyzed through a Voronoi diagram, which is a foundational algorithm that divides an image around a set of points into a series of convex regions. She had always been drawn to the structured yet organic feel of the technique.

Xie experimented with passing the distorted character path points as inputs into the Voronoi diagram. She played with separating the shapes created by the background versus the inner shapes of the characters—a step that gave her flexibility to further manipulate the Seal Script.

To Emily, the lines she had created began to resemble broken glass. She found significant beauty in the shattering of these shapes.

The Process

Phase II: Stitching Together

Seeking to incorporate textiles, a key area of exploration in her practice, Emily selected tapestry images from the public domain via Metropolitan Museum of Art’s open access collection and imported Chinese embroidery textures into the shapes to fill them.

Continuing to further fragment the images and textures, Emily discovered something interesting about the recursiveness: something that she wanted to mirror in the patterns within the characters. She then programmed into her algorithm a procedure that randomly selects a portion of the tapestry for each inner fragment, shrinking down, or blowing up the dimensions depending on the fragment and clipping them into the convex shapes.

Emily then introduced a series of glitch effects. Over time, she had begun to slowly incorporate glitching into her practice: she had for so long thought of computation as a way to create out of nothing, but has more recently found fascination with these techniques as they represent a tradition of destroying through means of computing.

Exploring this space between composition and destruction eventually led Emily to a fork in the road: a point at which she had the choice between two directions –a more minimalist versus maximalist approach. There was more at stake in this decision than aesthetic differences, however, prompting additional reflection: What do these characters mean to Emily? How can Emily portray this relationship through shapes and textures? How will others interact with and derive their own meaning from Emily’s visual language?

Images of the final outputs of Xie’s algorithm

After much exploration, Emily chose the minimalist route, preserving the general shapes of the divs to a larger extent. Historically, her works have been characterised by a more maximalist visual approach. However, she felt that it was not appropriate in this case. Here, she was driven by a desire to keep the essence of the forms legible, even as they were fractured and reshaped. A minimalist approach in the shape’s transformations allowed her to tow a more nuanced line between transformation and continuity––as a way to underscore both the fragility yet resilience of these forms. She wanted to keep the fact that these were Seal script characters apparent, and she wanted the overall images to retain their vague resemblance to ancient scrolls.

Images of the final outputs of Xie’s algorithm

Ultimately, Xie observed how the fractured lines in her works began to coincidentally resemble thread, or stitching. Perhaps, rather than being ‘broken’ or ‘fractured’ as she had once suggested, Emily’s cultural ties might be read as sewn together: a metaphorical ‘quilt’ of stories and languages. Much like her own heritage, the seal script characters, though distorted and broken apart, retained their underlying divs and identities––a semblance of the past, carried into the present in altered form.

V: The MODEL

The result of this experiment is a model that conveys Emily’s relationship to her multifaceted cultural identity through her very own AI-driven language.

Xie was interested in how the model, when prompted with figurative elements, transformed the destroyed Seal Script into more representational forms. The TITLES team, recognizing this potential, helped to train the model to amplify this aspect of it. Xie found it fascinating to take a character set that marked a pivotal, evolutionary step toward the Chinese writing system that we know today, and use technology not to push it further into abstraction, but to pull back it into its roots of figuration.

Seal Script is interesting in the sense that it represents an inflection point in Chinese writing, as it bridges the abstract and representational. If one engages with the script through modern computational tools, we highlight not only the artistry, but also that connection between language, imagery, and historic cultural underpinnings. First, by destroying it using computation, and then by bringing new life and meaning to the fragments it leaves behind. Ultimately, Xie felt that this exploration investigates the boundaries of what machine learning and AI can underscore about the interplay of innovation and tradition.

This experiment, however, is ongoing. In addition to the layers of language that have been incorporated, we welcome your language – your prompts. This interaction adds yet one more layer to the depth in this work, and transforms it into an interactive dialogue between artist, viewer, and machine. Your prompts represent an opportunity to reinterpret Seal Script through your own lens, allowing you to reshape fragmented glyphs, and rediscover and reframe historical symbols, reflecting the ever-evolving nature of tradition.