My WordPress

Colour Correspondence is a cameraless photography series made without a camera, without a scene, and without a subject in the conventional sense. In its place: a correspondence between chance and a trained machine — one colour posed as a question, another returned as an answer, the exchange fixed permanently onto photosensitive paper.

Dataset

At the core of the work is a custom machine-learning model, trained not for broad utility but for a single, obsessive purpose — to inscribe how one colour subjectively corresponds to another. The model is trained on a dataset of over 400 colour pairings, each one shaped by the particularities of a single viewpoint against colour generated by chance, outside of the averaged consensus of a large, general-purpose model. In this way, the machine becomes less a tool of computation and more an extension of subjectivity: a mirror that has absorbed a way of seeing and reflects it back at scale.

The colour on the left, on these dataset pairings, is generated by chance where the colour on the right was selected to correspond.

Model

Machine learning is usually associated with generalisation — with the smoothing of individual perspective into statistical consensus. Large models, trained on vast and varied data, tend to flatten the idiosyncratic in favour of the broadly legible. A custom model inverts this tendency. Rather than diluting a viewpoint into a norm, it amplifies it. The subjectivity encoded in the training data is not absorbed and neutralised — it is concentrated and extended across every output.


Works

The process is deliberately direct. A colour is selected by chance. The model corresponds with absolute certainty. That response is used as light — projected into the analogue colour darkroom, where its chromatic output exposes photosensitive paper.