Edit Spillover as a Probe: Do Image Editing Models Implicitly Understand World Relations?

TL;DR

This paper investigates whether the unintended spillover in image editing models indicates true world understanding by systematically analyzing spillover types, creating a benchmark, and evaluating multiple models.

Contribution

It introduces EditSpilloverProbe, a framework and benchmark dataset to assess world knowledge in image editing models through spillover analysis.

Findings

Spillover rates vary significantly across models, from 3.49% to 11.46%.

Semantic spillover correlates with models' world understanding capabilities.

Semantic spillover area density decays exponentially with distance, indicating genuine understanding.

Abstract

Instruction-following image editing models are expected to modify only the specified region while keeping the rest of the image unchanged. However, in practice, we observe a pervasive phenomenon -- edit spillover: models alter semantically related but unspecified content outside the edit region. This raises a fundamental question -- does spillover reflect genuine implicit world understanding, or is it merely attention leakage? We propose EditSpilloverProbe, a systematic framework that repurposes edit spillover as a natural probe for world knowledge in image editing models. We introduce a spillover taxonomy (spatial, semantic, mixed, random), an automated detection-and-classification pipeline, and a benchmark dataset constructed from real-world Chinese text editing tasks, EditSpilloverBench. Systematic evaluation of 5 representative editing models reveals three core findings: (1) spillover rates vary dramatically across architectures, from 3.49% to 11.46%, with a 3.3x ratio; (2) absolute semantic spillover quantity reveals models' world understanding capability -- nano_banana produces the most semantic spillover (27.8 per image), while qwen_2511 has the most precise editing control but lower semantic spillover (16.3 per image), revealing a trade-off between editing control and world understanding; (3) spatial decay analysis shows spillover area density decays exponentially with distance, but the proportion of semantically relevant spillover remains constant (40%-58%), providing direct evidence that semantic spillover reflects genuine world understanding rather than spatial diffusion.