Sitemap
Medium Logo

What if identity is not a fixed property, but a process — one that can be distilled, disrupted, and examined?

This project tests that question by inserting AI as a distillation apparatus between a past self and a future self — and asking whether anything essential survives the passage.

Drifting Contours

Drifting Contours is an ongoing experimental project that questions the nature of selfhood and authorship through a cyclical process: training an AI on my own existing works, manually transcribing the generated images by hand, and feeding the results back into the AI as new training data.

The process resembles biological metabolism — each cycle producing something recognisably similar, yet subtly transformed. What appears to be repetition is, in fact, continuous mutation. This metabolic loop is repeated continuously, with curatorial interventions introduced over time to maintain generative diversity.

The contradiction at the heart of contemporary AI discourse

Many artists resist having their work ingested into public AI datasets, fearing the loss of their distinctive style — while simultaneously using AI privately as a kind of ghostwriter. This contradiction — protect my identity, but exploit the tool — reveals something worth examining: what exactly is being protected, and why does it feel so threatening to let it go?

A different premise

This project begins from a different position: a fundamental skepticism toward the notion that "individuality" or "personal style" exists as a stable, definable entity. If such certainty cannot be assumed, then the fear of losing it dissolves — and an experiment becomes possible.

Rather than placing AI outside the self as a convenient instrument, this project positions the boundary between artist and AI — and the very question of existence — as the subject of inquiry itself.

The question being tested

If something — call it residue, or trace — is involuntarily embedded in an artwork during its creation, can an AI-mediated distillation process make that something visible? Does it survive the loop? And does the loop itself remain constant — or does the act of curating, repeating, and returning subtly alter the conditions of the experiment each time?

The answer remains unknown — and may remain so.

On the choice of subject: the human face

The decision to work with portraiture was initially practical: training AI on nature and botanical imagery — rendered in a painterly rather than photographic style — produced structurally incoherent outputs unsuitable for this process. The human face generated stable, workable results.

But the choice is also conceptual. The human face is the motif most universally and intimately familiar. Even subtle deviations register as uncanny. If something is being distilled through this process, the face is the surface most likely to make it perceptible — to both the artist and the viewer.



Itsuki Mokuda

Itsuki Mokuda

Painter, for years. Since 2026, on loop: train AI on my work, generate, repaint by hand, retrain — calmly testing what's left of "me" that isn't the model.