PromptLA: Towards Integrity Verification of Black-box Text-to-Image Diffusion Models

2024-12-20

arXiv.org

AI Summary

Black-box text-to-image (T2I) diffusion models are vulnerable to malicious fine-tuning that evades safety guardrails, posing critical challenges for integrity verification. Method: We propose the first quantitative integrity verification framework for T2I models. Our approach models generative image feature distributions via KL divergence to design an auditable, post-processing-robust detection paradigm. We introduce Prompt Learning Automata (PromptLA) for efficient black-box prompt selection and integrate feature-space statistical modeling with query optimization to minimize detection overhead. Contribution/Results: Evaluated on four major models—including SDXL and FLUX.1—our framework achieves a mean AUC of 0.962 (improving over baselines by >0.2) while substantially reducing query cost. This work establishes the first quantifiable, reproducible technical standard for regulatory oversight and copyright litigation of AI-generated content.

Technology Category

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

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Despite the impressive synthesis quality of text-to-image (T2I) diffusion models, their black-box deployment poses significant regulatory challenges: Malicious actors can fine-tune these models to generate illegal content, circumventing existing safeguards through parameter manipulation. Therefore, it is essential to verify the integrity of T2I diffusion models. To this end, considering the randomness within the outputs of generative models and the high costs in interacting with them, we discern model tampering via the KL divergence between the distributions of the features of generated images. We propose a novel prompt selection algorithm based on learning automaton (PromptLA) for efficient and accurate verification. Evaluations on four advanced T2I models (e.g., SDXL, FLUX.1) demonstrate that our method achieves a mean AUC of over 0.96 in integrity detection, exceeding baselines by more than 0.2, showcasing strong effectiveness and generalization. Additionally, our approach achieves lower cost and is robust against image-level post-processing. To the best of our knowledge, this paper is the first work addressing the integrity verification of T2I diffusion models, which establishes quantifiable standards for AI copyright litigation in practice.

Problem

• Verifying integrity of black-box text-to-image diffusion models
• Detecting model tampering via KL divergence of image features
• Proposing PromptLA for efficient and accurate integrity verification

Innovation

• Uses KL divergence for model tampering detection
• Introduces PromptLA for efficient prompt selection
• Achieves high AUC in integrity verification