Story perspectives

AI Image Generators Default to Generic Visuals, Study Reveals

12/21/2025

26 5

Summary

Story summary

A Patterns study finds AI image generators, including Stable Diffusion XL and LLaVA, produce generic outputs despite vast visual data.
Researchers conducted a visual telephone game with iteratively generated images from shifting prompts to test creativity.
They observed outputs coalescing around 12 common motifs, described as "visual elevator music."
The findings suggest AI tends to default to familiar styles, reflecting human preferences embedded in training data.

1 of 1

Story summary

A Patterns study finds AI image generators, including Stable Diffusion XL and LLaVA, produce generic outputs despite vast visual data.
Researchers conducted a visual telephone game with iteratively generated images from shifting prompts to test creativity.
They observed outputs coalescing around 12 common motifs, described as "visual elevator music."
The findings suggest AI tends to default to familiar styles, reflecting human preferences embedded in training data.