Fragmenta Workshop with Misagh Azimi
December 7 | 10:00 - 17:00

Commercial generative music platforms promise to “unleash your creativity” and that “anyone can make music.” Underneath, they took apart millions of recordings and turned them into statistical patterns, without asking the people who made them, crediting them, or paying them. Misagh Azimi calls the result “Music Creation as a Service”: always available, but cut off from the work, the scenes, and the struggles that give music its meaning. It doesn’t have to be that way.
Join Machine Agencies for a hands-on workshop on Fragmenta (https://misaghazimi.com/fragmenta), an open-source desktop application created by Misagh. It runs the whole text-to-audio process in one interface, with no coding: building a dataset, training a model, generating sound, and performing live. Built on Stable Audio 3, it lets you train LoRA adapters on your own recordings, so the model learns your own sound. Everything runs locally and after a one-time download, no audio ever leaves your machine.
What we’ll cover:
The session opens with a critique of commercial generative AI: who it takes from, who profits, and why that matters for musicians. The base model will download while we talk. Then we will go through the full process step by step:
- building a dataset in the Fragmenta Dataset Workbench
- labelling it by hand or with the built-in auto-annotation
- training a LoRA adapter
- generating from the result
- Performance Mode, a four-channel sampler with MIDI learn, played live as an instrument
Who it’s for:
Creative practitioners, musicians, composers, sound designers, and sonic artists, at any level of technical experience. It suits people curious about what training a model on their own archive involves, and people who want to bring local generative audio into a live or studio practice.
What to bring:
A laptop with Python 3.11 installed, running one of:
– Windows or Linux: an NVIDIA GPU with 8 GB+ VRAM is highly recommended.
– Mac: an Apple Silicon model (M1 or later) running macOS 14+.
If your machine doesn’t meet these, come anyway: you’ll get access to a demo Hugging Face space.
Places are limited.
Registration is required:
ABOUT MISAGH AZIMI:
Misagh Azimi is an interdisciplinary artist and researcher working across composition, music production and creative digital technologies. He created and maintains Fragmenta. He composes mainly for stage and visual media, and performs and releases music through the independent label Warm Mud Records. He is a guest researcher at the Metacreation Lab for Creative AI in Vancouver, where he works on generative AI for creative uses, critical views of technology in music (especially labour, authorship and technofeudalism), and new tools for making and performing music. Since 2019 he has taught music and media at universities in Germany and New Zealand. He is completing a PhD at Victoria University of Wellington on bending text-to-audio generative AI.
More at misaghazimi.com.
ABOUT MACHINE AGENCIES:
Machine Agencies is an experiment between human and machine intelligences. We are a collection of researchers investigating artificial intelligence technologies, the culture of AI development, and AI’s social, political, and environmental consequences. We encourage cooperation and play, resisting the antagonism of more instrumental approaches of AI. Our members are working on fascinating projects that bridge the gaps between engineering, artistic creation, academic debate, policy development, and public discourse.
https://machineagencies.milieux.ca/
December 7, 2026
10 AM-5 PM
Speculative Life Commons EV 10.625
