Those were generated by an AI. I dont know about the others but Stable diffusion can do that using controlnet.

Now this might not be accurate but most of these images are made by asking it to generate cats (and of course you prompt for everything else: backgrounds and details). Then you would take a picture of the words you want, so like open up a text editor and type what you need+screenshot. The words pic goes into the ControlNet module and i think you set it to use the “Tile” controlnet or i feel like Lineart works too maybe.

Hopefully this is enough to build on and get to what you want

i tried to use the stable diffusion tools a while back - there’s a lot of work involved & my hardware isnt sufficient enough to do much. perhaps things are easier now

I use runpod to rent a graphic card using the stable diffusion fast template.

There’s a jupyter notebook that you run, it gives you an external link at the end. You need to make sure to dowload all controlnet models.

Then you need to download both checkpoint models and the qr monster controlnet model from civitai. For models, I suggest realistic vision. You can download these into the runpod environment with the runpod terminal and typing in wget (download link) -O (model name).

The checkpoints will go in the SD stable diffusion model folder, while at monster will go underneath where qr monster gets installed in its own model folder (I forgot where it gets installed, something like extensions)

On the civitai qr monster, there’s more info including settings. I usually use euler a, 50 steps, and controlnet 1.4 weight, starts at 0.3 and ends at 0.9