Saved in:
Bibliographic Details
Main Authors: Yi, Jung, Kim, Minjae, Cho, Paul Hyunbin, Jang, Wooseok, Yun, Sangdoo, Kim, Seungryong
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.22718
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911705097830400
author Yi, Jung
Kim, Minjae
Cho, Paul Hyunbin
Jang, Wooseok
Yun, Sangdoo
Kim, Seungryong
author_facet Yi, Jung
Kim, Minjae
Cho, Paul Hyunbin
Jang, Wooseok
Yun, Sangdoo
Kim, Seungryong
contents Autoregressive video diffusion models have enabled real-time, action-conditioned world generation. However, sustaining a persistent world, where revisiting a previously seen viewpoint yields consistent content, remains an open problem. Full KV-cache attention preserves this consistency but breaks real-time constraints: memory footprint and attention cost grow linearly with rollout length. Sliding window inference restores throughput but discards long-term consistency. We propose WorldKV, a training-free framework with two components: World Retrieval and World Compression. World Retrieval stores evicted KV-cache chunks in GPU/CPU memory and selectively retrieves scene-relevant chunks via camera/ action correspondence, inserting them back into the native attention window without re-encoding. World Compression prunes redundant tokens within each chunk via key-key similarity to an anchor frame, halving per-chunk storage to fit 2x more history under a fixed budget. On Matrix-Game-2.0 and LingBot- World-Fast, WorldKV matches or exceeds full-KV memory fidelity at roughly 2x the throughput, and is competitive with memory-trained baselines without any fine-tuning. Project Page: https://cvlab-kaist.github.io/WorldKV/
format Preprint
id arxiv_https___arxiv_org_abs_2605_22718
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle WorldKV: Efficient World Memory with World Retrieval and Compression
Yi, Jung
Kim, Minjae
Cho, Paul Hyunbin
Jang, Wooseok
Yun, Sangdoo
Kim, Seungryong
Computer Vision and Pattern Recognition
Autoregressive video diffusion models have enabled real-time, action-conditioned world generation. However, sustaining a persistent world, where revisiting a previously seen viewpoint yields consistent content, remains an open problem. Full KV-cache attention preserves this consistency but breaks real-time constraints: memory footprint and attention cost grow linearly with rollout length. Sliding window inference restores throughput but discards long-term consistency. We propose WorldKV, a training-free framework with two components: World Retrieval and World Compression. World Retrieval stores evicted KV-cache chunks in GPU/CPU memory and selectively retrieves scene-relevant chunks via camera/ action correspondence, inserting them back into the native attention window without re-encoding. World Compression prunes redundant tokens within each chunk via key-key similarity to an anchor frame, halving per-chunk storage to fit 2x more history under a fixed budget. On Matrix-Game-2.0 and LingBot- World-Fast, WorldKV matches or exceeds full-KV memory fidelity at roughly 2x the throughput, and is competitive with memory-trained baselines without any fine-tuning. Project Page: https://cvlab-kaist.github.io/WorldKV/
title WorldKV: Efficient World Memory with World Retrieval and Compression
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2605.22718