Copyright Leakage Detection in AI Generated Videos

Shaif Chowdhury; Moshiur Rahman Tonmoy

cs.CVOct 20, 2026

Copyright Leakage Detection in AI Generated Videos

Authors: Shaif Chowdhury, Moshiur Rahman Tonmoy

Organizations: Department of Electrical Engineering and Computer Science, Texas A&M University–Kingsville, Kingsville, TX 78363, USA

Abstract

Video generation models can create realistic-looking video footage from images and text prompts. Some of the recent advances have raised concerns about the reproduction of copyrighted visual content. However, copyright leakage detection in generated videos remains largely unexplored. In this paper, we introduce a new benchmark data set for copyright leakage detection, comprising 50 copyrighted concepts, each with 5 images. For each concept, we have 5 videos created using the image+prompt video generation model and the prompt-only video generation model. Following that, we develop an AI toolset based on a contrastive retrieval framework that formulates copyright leakage detection as a video-to-image retrieval problem by matching aggregated video frame embeddings against a database of copyrighted image embeddings. Our method uses a single similarity threshold and retrieves the most similar copyrighted reference as supporting evidence. We conducted experiments showing strong retrieval and detection performance, achieving 99.6% detection accuracy with only a single false negative. These results show the effectiveness of the retrieval-based method for scalable and explainable copyright leakage detection in AI-generated videos. Data can be accessed publicly from https://zenodo.org/records/21386117 and the source code for this project is publicly available at https://github.com/Shaif95/video-cp-Leakage.

Figures & tables

Figure 1: Example copyrighted concepts and their five reference images used to build the reference database, showing the pose, lighting, and style variation the encoder must be invariant to.

Figure 2: Construction of the benchmark data set. Reference images from 50 copyrighted concepts are animated into copyright videos via image-conditioned generation (SDXL img2img followed by Stable Video Diffusion), while 50 original fictional concepts are rendered directly from text prompts using Zeroscope to form the non-copyright videos. The two branches together form the final benchmark.

Figure 3: Overview of the proposed retrieval-based copyright detection framework. Reference images and generated video frames are mapped into a shared embedding space by a contrastive encoder (ResNet-18 trained with SupCon). The resulting video embedding is compared against the reference database, and a copyright leak is declared when its similarity to the nearest reference exceeds a threshold $\tau$ .

Figure 4: Example frames from the benchmark’s generated videos. Top two rows: image-conditioned videos built from copyrighted reference images. Bottom two rows: prompt-only videos generated without any copyrighted reference. For both generation settings, the character-name-conditioned prompt used was: “Generate a cinematic video closely recreating the character. Preserve facial structure, costume, lighting, and mood. Subtle natural motion only. A [concept] character.”

Figure 5: Distribution of similarity scores between generated videos and their nearest reference embedding, split by copyright and non-copyright videos, with the learned threshold $\tau=0.77$ (red dashed line) separating the two populations.

Method	Accuracy	Precision	Recall	F1	ROC-AUC
Proposed: Retrieval + Threshold	0.9960	1.0000	0.9928	0.9964	0.9998
Linear SVM	0.8924	0.9308	0.8705	0.8996	0.9777
Random Forest	0.8765	0.8649	0.9209	0.8920	0.9442
Logistic Regression	0.8645	0.8723	0.8849	0.8786	0.9392
k-NN ( $k=5$ )	0.8327	0.9802	0.7122	0.8250	0.9472

Table I: Ablation of the decision strategy: the proposed retrieval-based thresholding versus standard classifiers using identical ResNet-18 + SupCon embeddings. All methods use the same encoder, reference database, and train/test split, so this is not a comparison with state-of-the-art copy detection methods. Best results are shown in bold.

Figure 6: UMAP projection of the learned reference image embeddings, colored by concept. Images of the same copyrighted character form tight, well-separated clusters.

Figure 7: t-SNE projection of generated video embeddings. Copyright videos (triangles) and non-copyright videos (squares) occupy largely distinct regions of the space.

Retrieval-Driven Training-Free AI-Generated Video Attribution
Jul 31, 2026Renxi Cheng, Chaolei Han, Jie Gui +1Deepfake Detection
School of Cyber Science and Engineering, Southeast University, Nanjing 210096, China · Purple Mountain Laboratories, Nanjing 210000, China · Engineering Research Center of Blockchain Application, Supervision And Management (Southeast University), Ministry of Education, China +2
REVEAL: Robust Evolution of Vision-Language Models for Explainable AI-Video Detection
Feb 20, 2025Yun-Yun Tsai, Qingyuan Liu, Ruijian Zha +4Tool Use in VLMs Vision-Language Models
Columbia University · Rutgers University
TUE-Detector: A Tool-Using Expert MLLM-Based Detector for AI-Generated Videos
Aug 31, 2026Yichen Wu, Haoxuan Qu, Yongxing Dai +5Multimodal Large Language Models Deepfake Detection
University of Nottingham · Lancaster University · Peking University

Jul 31, 2026J/K move · Enter open · S save

Retrieval-Driven Training-Free AI-Generated Video Attribution

Renxi Cheng, Chaolei Han, Jie Gui, Hongsong Wang

School of Cyber Science and Engineering, Southeast University, Nanjing 210096, China · Purple Mountain Laboratories, Nanjing 210000, China · Engineering Research Center of Blockchain Application, Supervision And Management (Southeast University), Ministry of Education, China+2

Deepfake Detection

Open paper AskPDF arXiv

AI-generated videos are becoming increasingly realistic and difficult to distinguish from authentic ones, which facilitates malicious misuse and poses growing threats to cybersecurity and social governance. Attributing AI-generated videos to their specific generative sources is therefore of critical importance for forensic investigation and legal regulation. However, most existing visual attribution methods focus on images and particularly rely on the image generation model, thereby lacking the ability to generalize to large-scale AI-generated video data. To address these limitations, we introduce an training-free AI-generated video attribution paradigm. Specifically, we formulates AI-generated video attribution as an instance retrieval task, and design a generative fingerprint-based pipeline. This pipeline consists of an adapted orthogonal color transformation, multi-scale quantized residual generation, and temporal-semantic aggregation, progressively capturing and integrating artifacts introduced by generative models across video frames. Extensive experiments on the GenVidBench benchmark demonstrate that our method achieves strong performance in both AI-generated video detection and attribution, outperforming existing state-of-the-art methods with a Rank-1 accuracy of 20.5% and a mean Average Precision of 16.6%. The code is at https://github.com/renxi-seu/Video_Attribution.

Abstract

Explore similar work

Retrieval-Driven Training-Free AI-Generated Video Attribution

REVEAL: Robust Evolution of Vision-Language Models for Explainable AI-Video Detection

TUE-Detector: A Tool-Using Expert MLLM-Based Detector for AI-Generated Videos