In the past, for example, it would be able to envision simple things like a landscape, Dayma said. But little by little, he's done things such as fixing bugs and improving code, enabling it to get better at coming up with more complicated images, such as the Eiffel Tower landing on the moon.

"When the model started drawing that, I was very happy," he said. "But then people came up with things even more creative, and somehow the model reached a moment where it was able to do something that looked like what they asked for, and I think that was a turning point."

The images Craiyon generates are not nearly as realistic-looking as what DALL-E 2 or Imagen can come up with, but they're fascinating nonetheless: People tend to blur into objects, and images look fuzzy and at least slightly askew.

For now, Craiyon is mostly being used for fun by people like Laming — perhaps in part because its results are not nearly as crisp or photorealistic as the images you can get from DALL-E 2 or Imagen, but also because people are still trying to figure out what to do with it. (The Craiyon website currently runs ads to recoup costs for the servers that power the AI system, and Dayma said he's trying to figure out how to make money from it while also allowing people to play with it for free.)