To save parsing the links, that's "hacker", "slug", "bed", and "teleport", all of which have perfectly reasonable hits.
I literally can't think of a halfway plausible coloring page that isn't covered that way. (Kinda thought I'd win that on "teleport", but... yeah, nope. Slim pickings at one result, sure, but that's still one reasonable result!)
Of course, you're pretty much writing off copyright laws in the process. How much that bothers you is up to you.
One of the big problems is that GANs can't generate files at a high enough resolution to make great coloring pages.
I then tried to make a neural network that would generate SVGs directly, which turns out to be really tricky too.
One thing that might work is to generate low resolution pages with a GAN, and then use an image to svg converter to make high resolution versions... maybe I'll try that next :)