当匿名性失效:识别文本到图像排行榜背后的模型
When Anonymity Breaks: Identifying Models Behind Text-to-Image Leaderboards
对做生成模型评测与图像取证的研究者,这篇给出一个可复用的去匿名化流程:用图像嵌入空间中的模型级质心做最近质心分类,无需提示词控制或训练数据,可用于检测排行榜匿名性漏洞或构建溯源/水印防御基线。
原文依据 generations from each T2I model form distinctive clusters in the image embedding space
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论文本到图像(T2I)模型日益流行,在线 AI 生成图像中很大一部分由它们产生。为比较模型质量,基于投票的排行榜已成为标准做法,并依赖匿名化的模型输出来保证公平。在本工作中,我们表明这种匿名性可以被轻易打破。我们发现,每个 T2I 模型的生成结果在图像嵌入空间中形成独特的簇,从而无需提示词控制或训练数据即可实现准确的去匿名化。使用 22 个模型和 280 个提示词(15 万张图像),我们基于质心的方法取得了高准确率,并揭示了系统性的模型特有签名。我们进一步提出一个提示词级可区分性指标,并开展大规模分析,展示某些提示词如何能带来近乎完美的可区分性。我们的发现暴露了 T2I 排行榜中的根本性安全缺陷,并推动更强的匿名化防御。
英文原文摘要 902 字符
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.