{"slug": "densitykv-density-guided-kv-cache-compression-for-long-video-generation", "title": "DensityKV: Density-Guided KV Cache Compression for Long Video Generation", "summary": "Researchers propose DensityKV, a training-free KV cache compression method that limits redundant historical token accumulation in autoregressive video diffusion models, improving long-horizon consistency and generation stability while keeping historical storage bounded independently of rollout length. The method, validated across three video generation backbones, uses Soft-Riesz density to constrain neighborhood-density growth in per-head token-level KV banks.", "body_md": "arXiv:2608.27922v1 Announce Type: new\nAbstract: Autoregressive video diffusion models enable streaming generation through sliding-window attention, but each generated block is conditioned on previously generated content, causing appearance and motion errors to propagate recursively over time. Historical key-value (KV) memory preserves earlier subject and scene states and helps maintain long-horizon consistency. However, retaining every generated state creates a historical archive that grows continuously with the rollout, while recurrent states repeatedly add redundant coverage. To address this problem, we propose DensityKV, a training-free historical KV bank management strategy. DensityKV maintains a separate token-level KV bank for each attention head and measures local redundancy among the post-RoPE keys that directly parameterize attention routing using Soft-Riesz density. By constraining neighborhood-density growth after states enter the bank, DensityKV limits repeated historical accumulation while preserving coherent states from each completed generation block. Experiments across three autoregressive video generation backbones and multiple generation lengths show that, at the same upper bound on historical KV capacity, DensityKV improves long-horizon consistency and generation stability while keeping persistent historical storage bounded independently of rollout length.", "url": "https://wpnews.pro/news/densitykv-density-guided-kv-cache-compression-for-long-video-generation", "canonical_source": "https://arxiv.org/abs/2608.27922", "published_at": "2026-08-31 04:00:00+00:00", "updated_at": "2026-08-31 04:23:03.390921+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "generative-ai", "ai-research"], "entities": ["DensityKV", "Soft-Riesz"], "alternates": {"html": "https://wpnews.pro/news/densitykv-density-guided-kv-cache-compression-for-long-video-generation", "markdown": "https://wpnews.pro/news/densitykv-density-guided-kv-cache-compression-for-long-video-generation.md", "text": "https://wpnews.pro/news/densitykv-density-guided-kv-cache-compression-for-long-video-generation.txt", "jsonld": "https://wpnews.pro/news/densitykv-density-guided-kv-cache-compression-for-long-video-generation.jsonld"}}