Artists Resurrect Veo 1, Google's Dead AI Video Model, in San Francisco

Google shut down Veo 1, its first text-to-video model, after researchers used it to generate hundreds of thousands of six-second videos — and now three artists have given that discarded archive a second life at tiat gallery in San Francisco.

The exhibition, "Post Generation," organized by Tina Tarighian, a senior creative technologist at Google's Creative Lab, asks what happens to AI models once faster successors render them obsolete. Veo 1, developed in 2024 and released only for internal use before eventually leaking, could produce barely functional clips: a six-legged horse galloping across a wet field, a three-limbed hominid swimming in a pool, a staticky smudge that might be children playing in leaves.

Researchers generated the videos before Google shut the model down and superseded it with Veo 2 and Veo 3, both released publicly.

Why keep a dead model's outputs?

"Because these models are so resource intensive and because each new model is faster and more efficient, it's hard to justify keeping old models around. 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?"

Tarighian shared a portion of the archive with three artists:

• Choreographer Zelia ZZ Tan
• Creative engineer Karyn Nakamura
• Fiber artist Haley Mortin

From garbled glyphs to embroidered samplers

For Mortin, the invitation extended a practice that already incorporates AI-generated images into knit works. Her fascination with AI artifacts traces back to 2018, when she took a job as a data annotator after graduating college — the often-invisible labor through which workers tag images to teach AI models to "see."

"It was 2018, I had just graduated college, I didn't know what AI was, and I needed money," Mortin said. "I thought it was just a data entry job."

Over two years she solved CAPTCHA tests and labeled images by subjective categories such as perceived emotion. Examining the Veo 1 outputs, she focused on the strange text the model generated.

"It produced two flavors of nonsense, these glyph-like shapes or empty phrases," Mortin said. "These [visual] models don't learn text like an LLM [large language model] does, rather, they learn text as a texture."

That insight led her to the embroidered samplers through which many 18th-century women first encountered writing. Her work Soup Is Made of Water (10 Samplers) (2026) comprises ten knitted pieces in black and white yarn that capture Veo 1's garbled text: a looping scribble, a calligraphic attempt at "beauty" [b eurt], and a shape resembling a sigil tattoo. The piece sets the language-learning practices of 18th-century women against those of contemporary machines.

An altar to video technology

Karyn Nakamura combined several video technologies in Labor of Light (2016), an assemblage suspended from cables like an altar to the history of video production. A white device emits a beam whose flicker contains the analog signal of Quicker 'n a Wink (1940), the film explaining how MIT's Harold Edgerton developed high-speed photography for slow-motion footage.

The beam leads to a projector casting a collage of Veo 1 videos onto white cloth, each clip emerging from RGB noise — the initial field of noise from which generative AI images and videos are built. Beneath the cloth, toothpicks and putty hold printouts of landmark research papers behind diffusion and other techniques underpinning modern AI.

"I find the process of image making and the process of video and video systems incredibly beautiful, there's a lot of craft in there," Nakamura said.

She worries the pace of development makes these systems harder to understand, limiting creative agency. Her work often dwells in the glitches of earlier models such as Stable Diffusion 1.5. "Because I like Stable Diffusion 1.5, sometimes I think, 'Oh, am I being a Luddite?'" Nakamura mused. "Like, what is even progress right now?"

Dancing the generated body

Choreographer Zelia ZZ Tan confronted the same tension in a ten-minute dance performed at the exhibition's opening on October 8, combining choreography with animated avatars that capture how bodies move in Veo 1 videos.

"I searched the archive through movement rather than only through words when I started to realize that between generated bodies, physical bodies, and the computational body, there are a lot of similarities and also a lot of rules based on how they can perform," Tan said.

Dancing within the limitations of the generated body required vulnerability. "It's such an honor to have a body, to be able to be present there, and maybe I'm not showing the most trained technique, but to listen, to be hesitant on stage, to wait for something to happen – that's part of the power of being there as a performer," Tan said.

As generative models replace one another in relentless succession — the "Will Smith eating pasta" test has gone from monstrous morphing blobs to physically plausible pasta in three years — "Post Generation" preserves the artifacts of a machine no one can run again, positioning obsolescence itself as material for art.