I2VWM: Robust Watermarking for Image to Video Generation

2025-09-22

AI Summary

To address the challenges of source image traceability and insufficient robustness of existing unimodal watermarking methods in image-to-video (I2V) generation, this paper proposes I2VWM—the first cross-modal digital watermarking framework tailored for I2V models. Methodologically, we introduce the “diffusion distance” metric to quantify watermark persistence across temporal diffusion steps; design a video-simulated noise layer to enhance robustness during training; and incorporate an optical flow alignment module to enforce inter-frame consistency during inference. The entire framework is built upon an end-to-end differentiable diffusion model for joint watermark embedding and extraction. Experiments demonstrate that I2VWM significantly improves watermark resilience against generative distortions on both open-source and commercial I2V models, while preserving visual imperceptibility, strong generalization across architectures, and practical deployability.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Security and Privacy: Data transparency and provenance

Abstract

The rapid progress of image-guided video generation (I2V) has raised concerns about its potential misuse in misinformation and fraud, underscoring the urgent need for effective digital watermarking. While existing watermarking methods demonstrate robustness within a single modality, they fail to trace source images in I2V settings. To address this gap, we introduce the concept of Robust Diffusion Distance, which measures the temporal persistence of watermark signals in generated videos. Building on this, we propose I2VWM, a cross-modal watermarking framework designed to enhance watermark robustness across time. I2VWM leverages a video-simulation noise layer during training and employs an optical-flow-based alignment module during inference. Experiments on both open-source and commercial I2V models demonstrate that I2VWM significantly improves robustness while maintaining imperceptibility, establishing a new paradigm for cross-modal watermarking in the era of generative video. href{https://github.com/MrCrims/I2VWM-Robust-Watermarking-for-Image-to-Video-Generation}{Code Released.}

Problem

• Addresses misuse of image-to-video generation via robust watermarking
• Traces source images in generated videos across modalities
• Measures temporal watermark persistence in AI-generated content

Innovation

• Introduces Robust Diffusion Distance for temporal watermark persistence
• Uses video-simulation noise layer during training for robustness
• Employs optical-flow-based alignment module during inference