VideoEraser: Concept Erasure in Text-to-Video Diffusion Models

2025-08-21

AI Summary

Text-to-video (T2V) diffusion models risk generating non-compliant videos due to unauthorized privacy-sensitive, copyrighted, or harmful content in training data. To address this, we propose the first fine-tuning-free, plug-and-play concept erasure framework for T2V models. Our method operates in two stages: (1) Selective Prompt Embedding Adjustment (SPEA), which attenuates target concept representations within text encoder embeddings; and (2) Adversarial-Robust Noise Guidance (ARNG), which enhances denoising robustness against harmful semantic cues. The framework is model-agnostic and integrates seamlessly with mainstream T2V diffusion architectures. Evaluated across four erasure categories—objects, artistic styles, celebrities, and harmful content—it achieves a 46% average reduction in undesirable generations, outperforming all existing baselines and establishing new state-of-the-art performance. It further demonstrates strong generalization across diverse concepts and exceptional deployment efficiency.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Agentic search
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

The rapid growth of text-to-video (T2V) diffusion models has raised concerns about privacy, copyright, and safety due to their potential misuse in generating harmful or misleading content. These models are often trained on numerous datasets, including unauthorized personal identities, artistic creations, and harmful materials, which can lead to uncontrolled production and distribution of such content. To address this, we propose VideoEraser, a training-free framework that prevents T2V diffusion models from generating videos with undesirable concepts, even when explicitly prompted with those concepts. Designed as a plug-and-play module, VideoEraser can seamlessly integrate with representative T2V diffusion models via a two-stage process: Selective Prompt Embedding Adjustment (SPEA) and Adversarial-Resilient Noise Guidance (ARNG). We conduct extensive evaluations across four tasks, including object erasure, artistic style erasure, celebrity erasure, and explicit content erasure. Experimental results show that VideoEraser consistently outperforms prior methods regarding efficacy, integrity, fidelity, robustness, and generalizability. Notably, VideoEraser achieves state-of-the-art performance in suppressing undesirable content during T2V generation, reducing it by 46% on average across four tasks compared to baselines.

Problem

• Preventing harmful content generation in text-to-video models
• Removing unauthorized concepts from video diffusion outputs
• Blocking explicit and copyrighted material in AI videos

Innovation

• Training-free concept erasure framework
• Plug-and-play two-stage integration module
• Selective prompt embedding and noise guidance