The paper introduces V²A-Mark, a versatile deep visual-audio watermarking framework designed to address the challenges of multimedia forensics in the age of AI-generated content. V²A-Mark embeds invisible cross-modal watermarks into video frames and audio, enabling precise manipulation localization and copyright protection. The method combines the fragility of video-into-video steganography with robust deep watermarking, allowing for both visual and audio tamper localization and copyright extraction. Key contributions include: 1. **Design of V²A-Mark**: An innovative framework that embeds visual localization and copyright watermarks into video frames and audio samples, enabling precise manipulation localization and copyright protection. 2. **Temporal Alignment and Fusion Module (TAFM)**: Enhances temporal consistency and robustness by aligning supporting frames with reference frames. 3. **Degradation Prompt Learning (DPL)**: Improves robustness against common video and audio degradations by learning degradation prompts. 4. **Cross-Modal Extraction Mechanism**: Combines visual and audio information to extract final copyright information. 5. **Performance Evaluation**: V²A-Mark outperforms existing methods in localization accuracy, generalization, and copyright precision, as demonstrated on a visual-audio tampering dataset. The paper also discusses related work, method details, and experimental results, highlighting the effectiveness and advantages of V²A-Mark in various scenarios, including video and audio tamper localization and copyright protection.The paper introduces V²A-Mark, a versatile deep visual-audio watermarking framework designed to address the challenges of multimedia forensics in the age of AI-generated content. V²A-Mark embeds invisible cross-modal watermarks into video frames and audio, enabling precise manipulation localization and copyright protection. The method combines the fragility of video-into-video steganography with robust deep watermarking, allowing for both visual and audio tamper localization and copyright extraction. Key contributions include: 1. **Design of V²A-Mark**: An innovative framework that embeds visual localization and copyright watermarks into video frames and audio samples, enabling precise manipulation localization and copyright protection. 2. **Temporal Alignment and Fusion Module (TAFM)**: Enhances temporal consistency and robustness by aligning supporting frames with reference frames. 3. **Degradation Prompt Learning (DPL)**: Improves robustness against common video and audio degradations by learning degradation prompts. 4. **Cross-Modal Extraction Mechanism**: Combines visual and audio information to extract final copyright information. 5. **Performance Evaluation**: V²A-Mark outperforms existing methods in localization accuracy, generalization, and copyright precision, as demonstrated on a visual-audio tampering dataset. The paper also discusses related work, method details, and experimental results, highlighting the effectiveness and advantages of V²A-Mark in various scenarios, including video and audio tamper localization and copyright protection.