AI QR code art

The technique became widely known in 2023, when people used the Stable Diffusion image model with ControlNet, an add-on that steers generation with a control image, to paint scenes whose light and dark areas follow a QR code’s pattern. Newer image models can be given the code as a reference, and some tools blend the true pattern back into the finished image.

Whatever the method, a phone can read the result only if the same things survive that any QR code needs: three recognizable finder patterns, a clear difference in brightness between dark and light modules at their centers, a plain margin playing the role of the quiet zone, modules large enough at the scanning distance, and a code that is not bent out of its grid by perspective.

Many AI art codes look convincing and do not scan, or scan on one phone and fail on another. Error correction absorbs some of the art’s changes, but it cannot rebuild finder patterns, and a higher level adds modules, leaving less room for the picture. That is why lower levels with fewer, larger modules are sometimes easier to turn into art.

Two checks matter before printing. First, decode the image with a real decoder and confirm it returns the exact link. Second, print a proof at the real size and scan it with several iPhones and Android phones in the light where it will hang. Using a dynamic code inside the art also means the destination can change without commissioning a new image.