The Violence of AI Repair
10 April 2026
The unresolved is not an error of the image—it is what escapes its governance.
If the history of photography is inseparable from loss, it is because the photographic image never guaranteed the integrity of what it rendered visible. Blur, grain, overexposure, and absence were not simply technical limitations but conditions of appearance.
The photograph did not resolve the world; it registered its instability.
What appeared within the frame could withdraw, fragment, or remain indeterminate. This indeterminacy was not a failure of the medium but its epistemological force: the image held open the possibility that what is seen cannot be fully known.
Contemporary computational imaging systems operate according to an inverse logic.
Rather than recording the contingencies of the visible, they anticipate and correct them. Missing data is inferred, damaged surfaces are reconstructed, and visual noise is suppressed. The image is no longer the trace of an encounter but the product of a prediction.
In this regime, uncertainty is not preserved—it is eliminated.
What is commonly described as enhancement, restoration, or generation must therefore be reconsidered. These operations do not simply improve the image; they enforce a norm of visual coherence.
The algorithm does not ask what is there, but what should be there.
The result is not an image that reveals more, but one that deviates less.
In this sense, repair is not a neutral technical operation.
It is a form of governance.
To repair an image is to impose a horizon of acceptability—to determine the limits within which it can appear as intelligible.
The unresolved, the excessive, and the indeterminate are treated as errors.
The image becomes a site of compliance.
