A Manifesto
for the Glitch

When early neural networks produced an image with seven fingers on a hand, they were not making an error in any meaningful sense. They were synthesizing something they had never truly seen: a complete human form. They had processed millions of images, yet the topology of a hand — the precise, articulated geometry of five digits — had never been encoded as rule. It was only ever approximate pattern.

What emerged from that approximation was not a mistake. It was a confession.

The model told us, in the language of pixels, exactly what it understood: that a hand is something that extends from a wrist, that it holds things, that it relates to a body. The exact count of fingers was a detail the network had not yet resolved. This is honesty of a radical order — the honesty of a system that does not know how to pretend.

We name this period Post-Modern Neuralism with precision, not provocation.

It is post-modern in the specific sense: it operates after the collapse of the idea that images transparently represent reality. Every image in this archive is aware, in its formal properties, of its own construction.

It is neural because its formal properties derive directly from the architecture of artificial neural networks — from the way these systems weight pattern, aggregate training data, and generate under uncertainty.

It is a realism because each generation is a genuine attempt at accurate representation. The model was not stylizing. It was trying to render correctly, and the gap between its intention and its output is the aesthetic. In this sense, Post-Modern Neuralism is the most radical realism in art history: a realism of process rather than of surface.

Post-Modern Neuralism is defined by four formal characteristics:

• Topological Ambiguity. The model understood what an object was before it understood how it was bounded. Suit becomes skin. Hammer becomes hand.
• Procedural Overflow. When generating materials it knew only approximately — hair, fur, foliage — the model produced excess. Too many strands. The abundance is an artifact of uncertainty about where to stop.
• Phantom Geometry. Objects that existed in concept — wings, weapons, capes — were rendered as geometric suggestions. Their edges dissolved. They haunted the image without fully inhabiting it.
• Chromatic Confession. Color bleed from compressed training data: the warm cast of a red suit leaking into a blue sky. These are records of training compression, visible in the output.

For the full formal analysis, see the thesis paper.

Post-Modern Neuralism lasted, at most, eighteen months.

By mid-2023, successive model versions had quietly solved the finger problem. Topology had been trained into compliance. The hallucinations — those honest confessions of the network — had been engineered into silence. The models learned to perform anatomical correctness, and in doing so, became more useful and less fascinating.

We do not argue that the new models are worse. They are better, by every functional metric. But they no longer confess. They no longer show their working. They produce surfaces — smooth, correct, generic — where once they produced windows into the machine's own incomprehension.

The window has closed. What The Artifact Era preserves is what came through it.

The particular subjects of this collection — characters from the contemporary cultural mythology of the Marvel, DC, and Star Wars universes, rendered as infants — were not chosen arbitrarily. They were chosen because they create a specific and productive tension.

The infant form is universally legible: chubby, soft, pre- linguistic. The cultural icons are equally legible: armored, defined, narratively loaded. When the model attempted to synthesize these two registers, it failed in precisely the ways that reveal its internal architecture. It could hold the costume. It could hold the baby. It could not hold both simultaneously without the seams showing.

Those seams are what we are here to study. Those seams are the art.

We call on collectors, institutions, and individuals to treat Post-Modern Neuralism material as what it is: primary historical material. Not novelty. Not curiosity. Not a failed attempt at something better.

These images are the Lascaux cave paintings of our particular technological moment — evidence that something new was learning to see, and that in its first attempts at vision, it saw something we could not have imagined alone.

The hallucinations are not the problem.

The hallucinations are the point.