This shirt makes you invisible to AI cameras

A loud shirt turns out to be enough to disappear from an AI camera. Berlin artist Simon Weckert built one that object-detection systems fail to see, and he shot the demo at the exact station where Berlin is piloting the cameras.

How the pattern works

Weckert, known for creating a fake Google Maps traffic jam with a cart of 99 phones, designed Digital Camouflage in response to the AI video surveillance pilot at Berlin’s Kottbusser Tor. The pattern uses intense colors and overlapping shapes to disrupt the visual features detection models rely on to assemble a head, limbs and torso into one recognizable person. To human eyes it is simply a vivid shirt. To the model, the body stops adding up.

In his demonstrations with the open-source YOLO detection system, pedestrians around the wearer get boxed and labeled as persons while the wearer walks through unmarked.

What the project claims, and what it does not

Weckert is careful with the claim: the shirt is not guaranteed to defeat every surveillance system, and he frames the work as commentary on how easily automated vision can be disrupted. That framing is the sharp part. The infrastructure being rolled out to watch public spaces rests on models whose failure modes are physical, visible and reproducible with fabric.

As computer vision spreads through city surveillance, adversarial fashion has an obvious future as an actual category, and an equally obvious cat-and-mouse dynamic ahead: patterns tuned against today’s detectors will meet models trained on those patterns. The current version simply demonstrates that the mouse is, for now, ahead.