TwoHamsters: Benchmarking Multi-Concept Compositional Unsafety in Text-to-Image Models

TL;DR

This paper introduces TwoHamsters, a benchmark to evaluate multi-concept compositional unsafety in text-to-image models, revealing significant vulnerabilities in current safety mechanisms.

Contribution

The paper formalizes MCCU, creates a large benchmark dataset, and evaluates state-of-the-art models and defenses, exposing critical safety gaps in T2I models.

Findings

Current models like FLUX have a 99.52% success rate in MCCU generation.

Defense mechanisms like LLaVA-Guard only achieve 41.06% recall.

Existing safety measures are insufficient against MCCU vulnerabilities.

Abstract

Despite the remarkable synthesis capabilities of text-to-image (T2I) models, safeguarding them against content violations remains a persistent challenge. Existing safety alignments primarily focus on explicit malicious concepts, often overlooking the subtle yet critical risks of compositional semantics. To address this oversight, we identify and formalize a novel vulnerability: Multi-Concept Compositional Unsafety (MCCU), where unsafe semantics stem from the implicit associations of individually benign concepts. Based on this formulation, we introduce TwoHamsters, a comprehensive benchmark comprising 17.5k prompts curated to probe MCCU vulnerabilities. Through a rigorous evaluation of 10 state-of-the-art models and 16 defense mechanisms, our analysis yields 8 pivotal insights. In particular, we demonstrate that current T2I models and defense mechanisms face severe MCCU risks: on TwoHamsters, FLUX achieves an MCCU generation success rate of 99.52%, while LLaVA-Guard only attains a recall of 41.06%, highlighting a critical limitation of the current paradigm for managing hazardous compositional generation.