T2VAttack: Adversarial Attack on Text-to-Video Diffusion Models

TL;DR

This paper investigates the vulnerability of text-to-video diffusion models to adversarial attacks, proposing new attack methods that manipulate prompts to significantly degrade video quality and coherence.

Contribution

It introduces T2VAttack, a novel framework with semantic and temporal attack strategies, revealing critical weaknesses in state-of-the-art T2V models.

Findings

Minor prompt modifications can cause substantial degradation in video quality.

Adversarial attacks effectively disrupt semantic fidelity and temporal dynamics.

Current T2V models are highly vulnerable to simple prompt perturbations.

Abstract

The rapid evolution of Text-to-Video (T2V) diffusion models has driven remarkable advancements in generating high-quality, temporally coherent videos from natural language descriptions. Despite these achievements, their vulnerability to adversarial attacks remains largely unexplored. In this paper, we introduce T2VAttack, a comprehensive study of adversarial attacks on T2V diffusion models from both semantic and temporal perspectives. Considering the inherently dynamic nature of video data, we propose two distinct attack objectives: a semantic objective to evaluate video-text alignment and a temporal objective to assess the temporal dynamics. To achieve an effective and efficient attack process, we propose two adversarial attack methods: (i) T2VAttack-S, which identifies semantically or temporally critical words in prompts and replaces them with synonyms via greedy search, and (ii) T2VAttack-I, which iteratively inserts optimized words with minimal perturbation to the prompt. By combining these objectives and strategies, we conduct a comprehensive evaluation on the adversarial robustness of several state-of-the-art T2V models, including ModelScope, CogVideoX, Open-Sora, and HunyuanVideo. Our experiments reveal that even minor prompt modifications, such as the substitution or insertion of a single word, can cause substantial degradation in semantic fidelity and temporal dynamics, highlighting critical vulnerabilities in current T2V diffusion models.