preprint 2026 ยท arXiv

CLPIPS: A Personalized Metric for AI-Generated Image Similarity

Khoi Trinh, Jay Rothenberger, Scott Seidenberger, Dimitrios Diochnos, Anindya Maiti

arXiv

Abstract

This paper introduces CLPIPS, a customized variant of LPIPS designed to better align image similarity scoring with human judgments in iterative text-to-image regeneration workflows.

Key Results

  • Fine-tuned similarity scoring better matches human rankings than baseline LPIPS
  • Lightweight human-augmented tuning can materially improve perceptual alignment
  • Similarity metrics can serve as adaptive feedback tools in human-in-the-loop generation tasks