The Image Learns to Read Itself

I begin with drawings and mixed-media works made by hand: stains, fragments, leaves, diagrams, colour fields, and asemic marks. These images already contain rhythms and relations, but they do not arrive as a readable system.

I then submit each work to an AI-assisted transformation process. The machine is not asked to decorate, improve, or imitate the image. It is asked to treat the original as evidence of a latent visual language and to reorganize its marks through repetition, mutation, spacing, clustering, interruption, and failed alignment.

The transformed image becomes a second state: an artificial syntax derived from human gesture.

From this syntax, I extract a structural invariant—a relation that appears to govern the image. The invariant is then translated into a concise law, a speculative formula, and a minimal diagram. These forms are assembled into an Extracted Law Plate.

The work moves through three states:

Human Gesture → Artificial Syntax → Extracted Law

The process does not ask what the original image symbolizes. It asks what structural truth the image becomes capable of stating after translation.