AI as authorship: why I train my own models instead of using off-the-shelf tools

The difference between using AI and directing it. A reflection on custom model training as a form of artistic intent.

The distinction that matters

There is a fundamental difference between an artist who uses a pre-trained model as a filter and one who builds a model from their own data, tuned to a specific expressive purpose. The first is closer to a photographer choosing a lens; the second is closer to a photographer building their own camera. Both are legitimate — but they produce different relationships to authorship.

What custom training actually means

In my practice, custom training means collecting biometric data from specific sessions, annotating it according to my own emotional taxonomy, and training classifiers that reflect my understanding of the states I want to work with. The model does not know what inspiration means in general — it knows what inspiration looks like in my data, filtered through my choices about what to measure and how to label it.