Network Bending as Circuit-Bending Inspired Live Neural Synthesis Hacking

Błażej Kotowski, and Frederic Font

Proceedings of the International Conference on New Interfaces for Musical Expression

Abstract

Circuit bending has long framed technical malfunction and misuse as sites of musical invention, enabling non-expert, hands-on engagement with electronic sound circuits through the practice of an immediate canvas. This paper articulates Network Bending as a circuit-bending–inspired approach to neural audio synthesis, making its connection to the tradition and ethos of circuit bending explicit. Network Bending is presented here as realtime, hands-on modification of neural network parameters—specifically weights and biases—during sound generation, where instability, breakdown, and opacity become compositional material. The paper situates this practice within histories of musical hacking and instrument repurposing, and introduces an open-source Network Bending tool that enables parameter-level intervention in neural audio models within a realtime environment. Drawing on repeated use in performances and workshops, we introduce the Bending Log as a compositional technique for documenting and reusing exploratory interventions across models and performance contexts, and present a complementary Network Blending experiment that combines corresponding parameter spaces across models. We further reflect on workshop-based dissemination and community uptake, positioning Network Bending as an instrument-centered performance practice, a pedagogical tool, and an exploratory engineering approach for engaging with the internal dynamics of neural audio systems.

Citation

Błażej Kotowski, and Frederic Font. 2026. Network Bending as Circuit-Bending Inspired Live Neural Synthesis Hacking. Proceedings of the International Conference on New Interfaces for Musical Expression. DOI: 10.5281/zenodo.20784161 [PDF]

BibTeX Entry

@inproceedings{nime2026_49,
 abstract = {Circuit bending has long framed technical malfunction and misuse as sites of musical invention, enabling non-expert, hands-on engagement with electronic sound circuits through the practice of an immediate canvas. This paper articulates Network Bending as a circuit-bending–inspired approach to neural audio synthesis, making its connection to the tradition and ethos of circuit bending explicit. Network Bending is presented here as realtime, hands-on modification of neural network parameters—specifically weights and biases—during sound generation, where instability, breakdown, and opacity become compositional material. The paper situates this practice within histories of musical hacking and instrument repurposing, and introduces an open-source Network Bending tool that enables parameter-level intervention in neural audio models within a realtime environment. Drawing on repeated use in performances and workshops, we introduce the Bending Log as a compositional technique for documenting and reusing exploratory interventions across models and performance contexts, and present a complementary Network Blending experiment that combines corresponding parameter spaces across models. We further reflect on workshop-based dissemination and community uptake, positioning Network Bending as an instrument-centered performance practice, a pedagogical tool, and an exploratory engineering approach for engaging with the internal dynamics of neural audio systems.},
 address = {London, United Kingdom},
 articleno = {49},
 author = {Błażej Kotowski and Frederic  Font},
 booktitle = {Proceedings of the International Conference on New Interfaces for Musical Expression},
 doi = {10.5281/zenodo.20784161},
 editor = {Benedict Gaster and João Tragtenberg and Anna Xambó and Tom Mitchell},
 issn = {2220-4806},
 month = {June},
 note = {},
 numpages = {6},
 pages = {419--424},
 title = {Network Bending as Circuit-Bending Inspired Live Neural Synthesis Hacking},
 track = {Paper},
 url = {http://nime.org/proceedings/2026/nime2026_49.pdf},
 year = {2026}
}