Theme and variations 012123

DALL-E takes a text prompt and generates images based on the AI’s interpretation of that prompt. Having obtained such an image, DALL-E can then create further variations on it, but prompting is no longer available for such variations—the connection weights in the neural network are reassigned according to rules to which we have no access.

Here is a series of images generated from this prompt: “an image of a conundrum wading in the river of forgetfulness”. The central figure in the first generation looks vaguely like a musical instrument, although the others are obviously quite different.

Here are four variations generated from the first image above. DALL-E has really focused on the “violin” looking shape and created images that represent it variously. While contemplating these variations, it is useful to keep in mind the original prompt was for “a conundrum wading in the river of forgetfulness”. The mystery to me is why the neural network represented that prompt as it did for the first generation, vaguely like a violin. It seems more obvious that the set of variations below on that generation is more or less reinforcing elements already present in the original image