(un)Stable (dis)Connection

Misagh Azimi

(un)Stable (dis)Connection
Image credit: Misagh Azimi

Abstract:

(un)Stable (dis)Connection is an improvised performance that merges a fine-tuned text-to-audio latent diffusion model with traditional improvisational techniques in Ableton Live. Acting as a co-creative agent, the AI model instantly generates fresh audio fragments in response to short textual prompts provided by the performer. These fragments are seamlessly imported into Ableton using Max for Live, where they are manipulated and used to improvise in real-time. By fine-tuning an open-source text-to-audio model on the author’s own recordings, the piece achieves some degree of stylistic coherence without sacrificing unpredictability. (un)Stable (dis)Connection highlights the creative potential of ethically sourced, open-source AI in music performance, foregrounding human–machine synergy as both a technical and artistic frontier.