A Picture is Worth a Thousand Prompts? Efficacy of Iterative Human-Driven Prompt Refinement in Image Regeneration Tasks
Abstract
Evaluates the efficacy of iterative human-driven prompt refinement in AI image regeneration tasks. Through controlled experiments, we assess whether humans can effectively reverse-engineer prompts from generated images and iteratively improve them.
Key Results
- Iterative refinement improves regeneration quality but with diminishing returns
- Significant individual variation in prompt inference ability
- Structured prompting strategies outperform ad-hoc approaches