Video Unlearning via Low-Rank Refusal Vector

TL;DR

This paper introduces a training-free, weight-update method for removing unsafe concepts from video diffusion models, significantly reducing harmful content generation while maintaining quality and prompt alignment.

Contribution

It presents the first closed-form, training-free framework for concept removal in video diffusion models using a low-rank refusal vector and contrastive factorization.

Findings

Reduces unsafe video generations by 36.3% and 58.2% on benchmarks.

Maintains video quality and prompt alignment after concept removal.

Operates without retraining or inference overhead.

Abstract

Video generative models achieve high-quality synthesis from natural-language prompts by leveraging large-scale web data. However, this training paradigm inherently exposes them to unsafe biases and harmful concepts, introducing the risk of generating undesirable or illicit content. To mitigate unsafe generations, existing machine unlearning approaches either rely on filtering, and can therefore be bypassed, or they update model weights, but with costly fine-tuning or training-free closed-form edits. We propose the first training-free weight update framework for concept removal in video diffusion models. From five paired safe/unsafe prompts, our method estimates a refusal vector and integrates it into the model weights as a closed-form update. A contrastive low-rank factorization further disentangles the target concept from unrelated semantics, it ensures a selective concept suppression and it does not harm generation quality. Our approach reduces unsafe generations on the Open-Sora and ZeroScopeT2V models across the T2VSafetyBench and SafeSora benchmarks, with average reductions of 36.3% and 58.2% respectively, while preserving prompt alignment and video quality. This establishes an efficient and scalable solution for safe video generation without retraining nor any inference overhead. Project page: https://www.pinlab.org/video-unlearning.