USC Libraries Board of Councilers' Ahmanson Lab Project Spotlight

Ahmanson Lab AI + ART Competition

The Event

While text-to-image generators allow users to generate stunning images--from surreal and fantastical dreamscapes to photorealistic portraits or film stills--with simple text prompts, composing the right prompt has emerged as its own kind of craft. This competition was a way to showcase the human and artistic elements of prompt composition and to encourage discussion, in our community, around the social issues that arise with increased access to these new technologies.

The Competition

Players competed in two consecutive rounds, attempting to craft and alter prompts that built off of, and responded to, their opponents ideas while generating appealing, interesting, and engaging images.

Each round began with two nouns and one verb chosen by Lab staff.

The first player then had three minutes to craft a prompt around those words, generate a series of four initial images using Midjourney, and chose one image for upscale. The upscaled image represented the final image for the player's turn.

The second player then had 3 minutes to assess the image and alter the existing prompt to improve it in some fashion. How might they make the current image more interesting, realistic, or engaging? How might they make it more fantastical, grotesque, or bizarre? How might they take the current topic/prompt in a different direction entirely? In altering the prompt, the second player could add as much text as they like but could only remove 1/2 of the words already in the prompt - thus working with, while still altering, their opponents ideas and style.

The first player then had 3 minutes to repeat the process.

After 3 volleys (6 turns each), the audience voted on the best of the 6 images generated throughout the full round.

 

Contents of this path:

This page references: