What happens when you perform live with generative AI that you can only partly control?
NordiCHI 2026, Vaasa, October 2026
The workshop: The Dissonance Workshop: Live Performance and Generative Friction with AI was a full-day workshop at NordiCHI 2026, held at Academill, Åbo Akademi in Vaasa. It was organised by Danny Snow, Alexander Cohen, Enrique Encinas, Johanna Schütt, Joseph Lindley and Marguerite Barry.
The workshop brought together researchers, designers, artists and performers around a shared question: what happens to live work when generative AI is late, unexpected or out of step? Instead of treating these frictions as problems to fix, we explored them as a creative material. We moved from framing and introductions, through a creative collage making, to a life performance where we made sure “the dissonance was played, not hidden”.
My contribution
I facilitated a hands-on session in which the group worked with paper and scissors, making collages of their dissonances with generative AI: the moments when the tools frustrated, surprised or resisted them. While we cut and glued, we talked, and the conversation was recorded. The collages and the conversation became two layers of the performance.
See the examples of the collages created by the workshop participants below (click on one image to enter full screen mode)
The image: I performed the visuals live. I placed the collages under a webcam and fed the image through Stable Diffusion in real time, so what appeared on screen was a hallucination of our collages. I could move, swap and layer the paper, but the model kept reinterpreting it on its own. Control was shared between my hands and the machine.
The voice: Eugénie Desmedt trained a GPT-2 model on the transcript of our discussion. Its voice ran as an overlay through the performance: our own words, returned to us slightly wrong.
The screen recording of the performance: With a Fly, Fontana Club, Vaasa, 5 October 2026
Film put together with the resident students of craitr, CC BY-NC-SA 4.0. More at dissonance.craitr.no.
What I learned:
“Partial control didn’t feel like failure. It felt like a duet: the less I tried to steer, the more interesting the images became.”
I shared the control of the visuals with 2 others: Tao and controllers, programmed to steer the visuals
Picture: Workshop performers with Stable Diffusion still playing on the screen background.
See the gallery of pictures from the live performance in the Fontana club in Vaasa.
