Janelle Shane at AI Weirdness tested four algorithms to see if neural networks can generate valentines. Sure they can! For this project, she used the simple, hokey type of Valentines that children give their classmates. Since the algorithms already knew English and already knew that Valentines Day is a holiday, she only fed ten examples into their training programs.
Listen beautiful relax classics on our Youtube channel.
The original intent was to have the machines generate both a greeting and an image, but Shane knew from experience that even if these algorithms could handle it (which they couldn’t), the images would illegible and creepy. So she asked for a greeting plus a text description of an image to go with it. While the cards she fed into the learning program had images that supported the pun wording in the greeting, these algorithms didn’t quite get it. But they generated a greeting and a random description of an image, which Shane illustrated herself. See a bunch more of these neural network-generated valentines at AI Weirdness.