TPIPS: Text-Prompted Image Perceptual Similarity Metric Released

junyanz89 · x · 2026-08-18

CMU and Adobe researchers introduced TPIPS, a text-conditioned image perceptual similarity metric. Unlike traditional metrics (LPIPS, SSIM) that provide a single score, TPIPS uses text prompts to specify visual aspects (e.g., lighting, pose) for comparison. Code is available via pip install tpips.

Original post →

More from Multimodal

Multimodal channel →