A Picture is Worth a Thousand Prompts? Efficacy of Iterative Human-Driven Prompt Refinement in Image Regeneration Tasks

TL;DR

The paper addresses how humans can iteratively refine prompts to regenerate target images and whether image similarity metrics align with human perception in this workflow. It combines a structured in-person study with subjective similarity rankings and ISM scores to quantify improvements across 10 iterations per image. The key findings show that early prompt refinements reliably improve alignment with the target, with diminishing returns later, and that some ISMs moderately reflect human judgments while others do not. The work demonstrates the potential and limits of ISMs as objective feedback in human-guided generative workflows, informing future human-AI collaboration tools and educational applications.

Abstract

With AI-generated content becoming ubiquitous across the web, social media, and other digital platforms, it is vital to examine how such content are inspired and generated. The creation of AI-generated images often involves refining the input prompt iteratively to achieve desired visual outcomes. This study focuses on the relatively underexplored concept of image regeneration using AI, in which a human operator attempts to closely recreate a specific target image by iteratively refining their prompt. Image regeneration is distinct from normal image generation, which lacks any predefined visual reference. A separate challenge lies in determining whether existing image similarity metrics (ISMs) can provide reliable, objective feedback in iterative workflows, given that we do not fully understand if subjective human judgments of similarity align with these metrics. Consequently, we must first validate their alignment with human perception before assessing their potential as a feedback mechanism in the iterative prompt refinement process. To address these research gaps, we present a structured user study evaluating how iterative prompt refinement affects the similarity of regenerated images relative to their targets, while also examining whether ISMs capture the same improvements perceived by human observers. Our findings suggest that incremental prompt adjustments substantially improve alignment, verified through both subjective evaluations and quantitative measures, underscoring the broader potential of iterative workflows to enhance generative AI content creation across various application domains.

Figures (7)

• Figure 1: Image generations using DALL-E 3 with prompts containing the same subject (dog) and different combinations of two modifiers (oil painting and bright colors).
• Figure 2: Summary of the iterative prompt refinement process for image regeneration task.
• Figure 3: Demographic distribution of survey participants.
• Figure 4: Participants' familiarity levels with different generative AI tools.
• Figure 5: An example of the iterative prompt refinement task with the ISM score feedback. The participant submits their prompt in the box, they will then see the generated image and the associated similarity score above. A green similarity score means the current iteration is better than the last, while a red similarity score means the opposite. To the left of the target image is an area to keep track of their best effort, including their best prompt and best similarity score.