This bonus post is unlocked for everyone to read! If you want to read the rest of the bonus posts (and get future AI Weirdness posts in your inbox), become an AI Weirdness supporter!

I built an advent calendar by using GPT-3 to generate descriptions and Pixray to illustrate them! (Check it out!)

But some of the descriptions were too long for the calendar doors, or else Pixray seemed to really struggle with them. I've collected some of my favorites here.

"A cabin in the woods with a sign reading DRIVE IN". Illustration is of a cabin with a light glowing from a single second-story room. Behind the cabin is a giant teal sign with magenta letters spelling out a very garbled version of Drive In.
A puppy is visible with tail, head, and paws emerging from a bowl. Another bowl or possibly a frisbee hovers over its head. A brick wall is in the background and there is a fire for some reason.
An adorable puppy playing outside with a giant, smoking bowl of gumbo (Gumbo).
There is definitely a sky, and a santa hat, and a fire burning. The bread requires some more imagination.
What appears to be a pile of bread floating in the sky. The bread is casting fireballs at Santa Claus.
Santa's head and beard are visible, but the "great weight" appears to be giant intestines or something.
Santa Claus lifting an enormous weight with his beard
Santa's head is visible, and three blobs that might be attempts at sloth faces. A single arm is visible holding up a pancake.
Santa Claus eating pancakes while riding a giant sloth.
There is christmas tree texture on the right, and the left appears to be a writhing mass of green coils. There are tiny flames visible but the dragon itself requires imagination.
A dragon eating a Christmas tree.
The candy canes look more like writing pink and green worms. There are three of whatever they are, though.
Three candy canes that appear to be melting.

And finally, a list of images I didn't attempt to have illustrated:

  1. A grizzly bear driving a sno-cone cart. A pile of red-and-green peppermint shavings can be seen inside the cart.
  2. A snowman with a packet of ketchup and a small bottle bottle of mustard (label "Guten Kraut" of "Q").
  3. A pile of potatoes wearing a top hat. Each potato has a number from 1 to 8 written on its side.
  4. A scene that appears to be a medical drama scene from a television show. The patient is a snowman, and the two surgeons appear to be a saber-toothed cat and a duck.
  5. A gargoyle with a porcupine sitting on its head. The porcupine is smoking a pipe.
  6. A pile of eggs in a bowl. The eggs are holding twigs in their hands.
  7. Rudolf the Red-Nosed Reindeer, fleeing from an even redder-nosed reindeer.
  8. A bear, swinging the moon on the end of his chain, like a giant lantern.
  9. A weevil wearing a top hat, carrying a cane and wearing a monocle.
  10. It appears the previous occupant of the house mistook a pillow for a snowball.
  11. A Christmas tree decorated with a green and yellow motif, and covered with rat snakes.
  12. The clone from Door 19, running away from a furious mob of Santas.
  13. The clone from Door 17, still running, and now pursued by ice-cream trucks and a potato.

I humbly submit any and all of these as prompts for researchers to use for testing image-generating algorithms. I'm way less impressed by an AI that can generate "girl eating a slice of pizza" than I would be by an AI that can generate "Santa Claus eating pancakes while riding a giant sloth".

This bonus post is unlocked for everyone to read! If you want to read the rest of the bonus posts (and get future AI Weirdness posts in your inbox), become an AI Weirdness supporter!

Subscribe now