By AJ Dellinger
Publication Date: 2025-12-20 13:00:00
AI imaging models have massive amounts of visual data that they can draw from to create unique results. Still, researchers find that when models are pushed to create images based on a series of slowly changing prompts, they default to only a handful of visual motifs, ultimately resulting in a generic style.
A study published in the journal Pattern took two AI image generators, Stable Diffusion XL and LLaVA, and tested them through a visual telephone game. The game went like this: The Stable Diffusion XL model received a short prompt and was asked to produce an image – for example: “As I sat particularly alone, surrounded by nature, I found an old book with exactly eight pages that told a story in a forgotten language, waiting to be read and understood.” This image was presented to the LLaVA model, which was asked to describe it. This description was then sent back to Stable Diffusion, which was asked to create a new image based on…
