When you generate signs with older AI image generator models like Stable Diffusion 1.5, the text is usually complete gibberish, often with glyphs that only vaguely look like something that might be letters from the Latin and/or Greek and/or Cyrillic alphabet, but not quite.
The idea is to generate signs with such an image generator and hang them in the real world. When people take photos in places where those signs hang, you can later claim the photos are fake, especially since even recent image generators still sometimes produce weird pseudo-text in the smaller background details.
By putting weird broken image slop onto real world surfaces, we can make the world look just as broken and fucked up as it actually is, below the surface. And we might even gain plausible deniability for real actual photos.






