Human Learning by Model Feedback: The Dynamics of Iterative Prompting with Midjourney

TL;DR

This paper investigates how human users iteratively refine prompts when generating images with Midjourney, revealing convergence patterns and potential biases towards the model's preferences that impact future training and user alignment.

Contribution

It provides the first dataset and analysis of iterative human prompts in text-to-image generation, highlighting convergence dynamics and biases in user-model interactions.

Findings

Prompts tend to converge toward specific traits during iterations

Both user realization of missed details and adaptation to model preferences influence convergence

Biases in prompts may reflect the model's preferences rather than human intent

Abstract

Generating images with a Text-to-Image model often requires multiple trials, where human users iteratively update their prompt based on feedback, namely the output image. Taking inspiration from cognitive work on reference games and dialogue alignment, this paper analyzes the dynamics of the user prompts along such iterations. We compile a dataset of iterative interactions of human users with Midjourney. Our analysis then reveals that prompts predictably converge toward specific traits along these iterations. We further study whether this convergence is due to human users, realizing they missed important details, or due to adaptation to the model's ``preferences'', producing better images for a specific language style. We show initial evidence that both possibilities are at play. The possibility that users adapt to the model's preference raises concerns about reusing user data for further training. The prompts may be biased towards the preferences of a specific model, rather than align with human intentions and natural manner of expression.