Digital Artists

An Artist Was Given the Leftover Archive of a Dead AI Model. She Knitted It Into Embroidery Samplers.

When Google shut down Veo 1, its early, barely-functional video model, a senior creative technologist asked three artists what to do with the hundreds of thousands of leftover generations. Fiber artist Haley Mortin turned the model's garbled on-screen text into knitted 18th-century-style samplers.

Most defunct AI models just disappear — the weights get taken offline, the outputs they generated vanish with them, and nobody mourns a superseded checkpoint. When Google shut down Veo 1, its earliest and least stable video-generation model, Tina Tarighian, a senior creative technologist at Google’s Creative Lab, had a different idea: give the leftover archive of outputs to artists and see what they made of it. The resulting show, Post Generation, paired three artists with a portion of that archive. The most striking response came from fiber artist Haley Mortin, who used it to knit embroidery samplers.

What was actually in the archive

Veo 1 was released only for internal use before eventually leaking, and by most accounts it barely worked: it could produce clips only six seconds long, with long waits between generations and results that were reliably, predictably strange — a six-legged horse galloping across a wet field, a staticky smudge that might be children playing in a pile of leaves, a three-limbed figure swimming in a pool. Researchers generated hundreds of thousands of these clips before the model was retired and replaced by Veo 2 and Veo 3, both of which were released publicly. “No one can generate using Veo1, and the outputs it produced aren’t publicly available,” Tarighian said. “So I thought, what can we do with these leftover scraps of a model? Can we gift this archive to artists?” She shared a portion of it with Mortin, media artist Karyn Nakamura, and choreographer Zelia ZZ Tan.

From data annotation to embroidery

Mortin’s fascination with AI artifacts isn’t abstract — it comes from having worked the other side of the pipeline. After graduating college in 2018 and needing money, she took what she thought was a simple data-entry job: solving CAPTCHAs and labeling images for two years, categorizing not just objects but subjective qualities like the emotion a photographed person seemed to be expressing. It’s the kind of invisible human labor that teaches AI models to see, and Veo 1’s outputs are, indirectly, downstream of exactly that work.

Looking through the archive, Mortin zeroed in on the garbled, glyph-like text the model generated whenever it attempted writing. “These [visual] models don’t learn text like an LLM does,” she observed, “rather, they learn text as a texture.” That observation led her to 18th-century embroidery samplers — the pieces young women once stitched to practice writing before they had regular access to ink and paper, often using them to record family histories in thread. The resulting work, Soup Is Made of Water (10 Samplers), comprises ten knitted pieces in black and white yarn, each capturing a different flavor of Veo 1’s nonsense text: a looping scribble, a calligraphic attempt at spelling “beauty,” and shapes that resist being read at all. It links two very different histories of text production taught through the body rather than through language — 18th-century needlework and 21st-century model training — and asks what gets preserved, and what gets lost, in both.

The rest of the archive, in other hands

Mortin’s work wasn’t the only response. Nakamura’s Labor of Light suspended a light-emitting device whose flicker secretly encoded a 1940 educational film about Harold Edgerton’s high-speed photography, with the beam projecting a collage of Veo 1 clips onto cloth alongside printouts of landmark AI research papers — a kind of altar to the history of image-making technology. Tan, for her part, performed a ten-minute dance built from studying how bodies move and glitch inside Veo 1’s generations, treating the archive as choreography rather than footage. Together, the three pieces turned a technically unremarkable, long-obsolete model into source material for a genuinely original meditation on what gets thrown away when a company moves on to the next release.