There are various tutorials, but my method is a bit different but it’s all basically the same idea:

• Download automatic1111

• Find a good stable diffusion model (comes with SD1.5), I’ve been playing with Reliberate recently

• install ControlNET extension

• generate QR code (use high error rate, like 30%)

• Use txt2img with QR code in ControlNet on maybe a “start at 0.21, end at 1” weight. (use tiling for best, imo)

• I use high sampling numbers too, like 40 or 50

• Try to get a prompt that’s friendly to tons of variation

• Generate a bunch, pick the best (you might have one that works already)

• Send back through img2img with the QR again in ControlNet

I made one real fast that's not super great but just to show example of one done on the fly. The secret is to try a bunch of variations and different models until you get stuff that looks good.

Another good method is to actually have a generated image (like your city or whatever) in ControlNet first run (use depth or canny or something) and use the prompt in a txt2img, but have the second unit ControlNet be the QR code (use tiling). It works good too and produces slightly different aesthetic.