This research introduces Cross-Modality Video Coding (CMVC), a groundbreaking paradigm that leverages Multimodal Large Language Models to achieve more efficient video compression.

• Integrates external knowledge priors from MLLMs into traditional video compression pipelines
• Disentangles video content on the encoder side for compact representation
• Employs generative models to reconstruct high-quality video on the decoder side
• Creates a unified framework that transcends traditional redundancy elimination approaches

This engineering innovation could significantly reduce bandwidth requirements for video streaming while maintaining visual quality—critical for evolving multimedia applications in bandwidth-constrained environments.

When Video Coding Meets Multimodal Large Language Models: A Unified Paradigm for Video Coding