Every time you see a stuffed animal with the wrong number/arrangement of limbs, take a shot

Poking an AI art bot, trying to get some scenery for a piece where I don’t feel like drawing a background from scratch. (It’s fanart, it’s not getting sold anywhere — my alternate strategy is using a screenshot from the show, which I would have equally-little copyright to.) (ETA: also, it's a free bot, no AI company is profiting either.)

Fascinating which prompt details it manages to follow, and which ones it seems to have no context for.

Example: it will render a passable “bedroom” setting…in broad daylight! At the angle you would expect in the average real-estate ad, or hotel photo!

But it full-on refused to produce one at night with the lights off. (References excised from the training data for being NSFW, maybe…?)

Sample of the bot’s efforts at a series of prompts:

 

( sitcom style, boy’s bedroom, unmade beds, overhead angled view, night lights off )