Identifying Models Behind Text-to-Image Leaderboards

2026-01-14

AI Summary

Current leaderboards for text-to-image models rely on anonymized outputs to ensure fairness, yet their anonymization mechanisms harbor critical security vulnerabilities. This work is the first to reveal that different models exhibit systematic and identifiable “signatures” in the image embedding space, enabling high-accuracy de-anonymization through clustering and centroid matching—even without access to prompts, control signals, or training data. We validate our approach across 22 state-of-the-art models and 280 prompts, generating a dataset of 150,000 images; under certain prompts, model identification accuracy approaches perfection. To quantify this phenomenon, we introduce the notion of “prompt-level distinguishability,” which exposes a fundamental flaw in existing anonymization protocols and challenges the assumption of model indistinguishability in benchmark evaluations.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness

Application Category

• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web search models and ranking

Abstract

Text-to-image (T2I) models are increasingly popular, producing a large share of AI-generated images online. To compare model quality, voting-based leaderboards have become the standard, relying on anonymized model outputs for fairness. In this work, we show that such anonymity can be easily broken. We find that generations from each T2I model form distinctive clusters in the image embedding space, enabling accurate deanonymization without prompt control or training data. Using 22 models and 280 prompts (150K images), our centroid-based method achieves high accuracy and reveals systematic model-specific signatures. We further introduce a prompt-level distinguishability metric and conduct large-scale analyses showing how certain prompts can lead to near-perfect distinguishability. Our findings expose fundamental security flaws in T2I leaderboards and motivate stronger anonymization defenses.

Problem

• text-to-image models
• leaderboards
• anonymity
• model identification
• security flaws

Innovation

• model deanonymization
• text-to-image models
• embedding clustering
• prompt distinguishability
• AI leaderboard security