Generative Video Compression: Towards 0.01% Compression Rate for Video Transmission

TL;DR

Generative Video Compression (GVC) tackles the problem of transmitting video at extreme compression rates by leveraging decoder-side generative priors to synthesize content from compact representations, achieving around to near $0.02\%$ compression and approaching $0.01\%$ in some scenarios. It introduces the AI Flow framework and views video transmission as Level C task-oriented communication, balancing bitrate, perceptual quality, and downstream task utility. The encoder outputs compact latent tokens, while a diffusion-based decoder reconstructs high-quality video from these tokens, guided by learned priors. Empirical results on the MCL-JCV dataset show perceptual quality at $0.008$ bpp with LPIPS around $0.214$, competitive downstream VOS performance on DAVIS2017, and practical latency on consumer-grade GPUs, illustrating the approach’s potential for bandwidth-constrained video communication.

Abstract

Whether a video can be compressed at an extreme compression rate as low as 0.01%? To this end, we achieve the compression rate as 0.02% at some cases by introducing Generative Video Compression (GVC), a new framework that redefines the limits of video compression by leveraging modern generative video models to achieve extreme compression rates while preserving a perception-centric, task-oriented communication paradigm, corresponding to Level C of the Shannon-Weaver model. Besides, How we trade computation for compression rate or bandwidth? GVC answers this question by shifting the burden from transmission to inference: it encodes video into extremely compact representations and delegates content reconstruction to the receiver, where powerful generative priors synthesize high-quality video from minimal transmitted information. Is GVC practical and deployable? To ensure practical deployment, we propose a compression-computation trade-off strategy, enabling fast inference on consume-grade GPUs. Within the AI Flow framework, GVC opens new possibility for video communication in bandwidth- and resource-constrained environments such as emergency rescue, remote surveillance, and mobile edge computing. Through empirical validation, we demonstrate that GVC offers a viable path toward a new effective, efficient, scalable, and practical video communication paradigm.

Figures (3)

• Figure 1: Overview of Our GVC Framework Grounded in the Shannon-Weaver model shannon1948mathematical. Top-left: Level A addresses the technical problem, optimizing signal fidelity under limited bandwidth by minimizing distortion between input and output videos. Top-right: Level B focus on the semantic problem, aiming at transmitting the precise semantic symbols. Bottom: Level C, central to the proposed Generative Video Compression (GVC) framework, emphasizing task-oriented effectiveness. It ensures that the compressed tokens enable the achievement of task goals - such as high-quality perception reconstruction or support for downstream tasks like segmentation.
• Figure 2: Bandwidth comparison for achieving comparable reconstruction quality. Traditional methods require more than 6 times the bandwidth to match the perceptual quality of our approach across selected representative sequences.
• Figure 3: Visual quality comparison of the miniaturized model, demonstrating competitive perceptual quality despite model compression.