The Database of Every Picture Imaginable

DALLE-2 Prompt: a photorealistic round glass fishbowl containing a collection of small colorful 2D images of people, places and things each using a random style, contrasting colors and themes

I wrote the post below in 1993 for April Fools, soon after I’d joined the Disney VR Studio. I posted it anonymously to USENET (the plaintext Reddit of the day).

The joke started with the realization that any image is just a unique number in the space of all possible images. If you string all the color bits of all the pixels together and read it sideways, it’s a really big number. It’s a number that may take as many bytes to represent as the image itself. So the Database of Every Picture Imaginable just needs to input this “image number” to find the image. Yeah, it’s that dumb. That’s why it’s an April Fool’s joke.

Today, you could think of the index as a text prompt that could be used to regenerate the desired image in a generative AI system. The AI Model is optimizing and effectively compressing all training data into a large plinko machine that can reconstruct any of it given the right prompt. And research indicates that you can recover most of the input images, lossily, with the right prompt.

Until now, 3D graphics pipelines have been the best form of large-scale image decompression, taking smaller 3D models (not AI models yet) and generating amazingly detailed 3D graphics from almost any angle. But I digress…

I’ve kept the typos and bad editing as is.