Quant VideoGen: Auto-Regressive Long Video Generation via 2-Bit KV-Cache Quantization

TL;DR

Quant VideoGen introduces a novel KV cache quantization framework that significantly reduces memory usage in autoregressive video generation, enabling longer, high-quality video synthesis on standard hardware.

Contribution

The paper proposes a training-free, multi-stage residual quantization method leveraging video redundancy, improving memory efficiency and generation quality in autoregressive video diffusion models.

Findings

Reduces KV cache memory by up to 7.0 times

Achieves less than 4% latency overhead

Outperforms existing baselines in quality on multiple benchmarks

Abstract

Despite rapid progress in autoregressive video diffusion, an emerging system algorithm bottleneck limits both deployability and generation capability: KV cache memory. In autoregressive video generation models, the KV cache grows with generation history and quickly dominates GPU memory, often exceeding 30 GB, preventing deployment on widely available hardware. More critically, constrained KV cache budgets restrict the effective working memory, directly degrading long horizon consistency in identity, layout, and motion. To address this challenge, we present Quant VideoGen (QVG), a training free KV cache quantization framework for autoregressive video diffusion models. QVG leverages video spatiotemporal redundancy through Semantic Aware Smoothing, producing low magnitude, quantization friendly residuals. It further introduces Progressive Residual Quantization, a coarse to fine multi stage scheme that reduces quantization error while enabling a smooth quality memory trade off. Across LongCat Video, HY WorldPlay, and Self Forcing benchmarks, QVG establishes a new Pareto frontier between quality and memory efficiency, reducing KV cache memory by up to 7.0 times with less than 4% end to end latency overhead while consistently outperforming existing baselines in generation quality. Code is available at: https://github.com/svg-project/Quant-VideoGen