Velox is a training‑free attention router that improves the efficiency of video world models by selecting region‑level context through frustum slab correspondences derived from streaming geometry. It represents past blocks as camera frustum slabs with depth intervals and uses a recency‑priority multiwarp to obtain current‑block depth, soft‑modulating importance for well‑covered regions while preserving model‑derived routing for uncovered areas. The fixed token‑block granularity aligns with FlashAttention‑style tiled execution, reducing the attention term that grows with rollout length and achieving significant speedups—4.36× acceleration in attention and 2.43× in denoising at chunk 48—while maintaining performance close to dense models on the WBench navigation split.
arXiv:2608. 12032v1 Announce Type: cross Abstract: Video diffusion transformers are costly to sample: every denoising step applies self-attention over a long 3D token sequence, a quadratic cost that dominates as resolution and duration grow.
arXiv:2608.23565v1 Announce Type: new Abstract: An interactive world model must follow the user's actions, remember the places it has shown, and stream in real time. The tension is structural: contro...
arXiv:2602. 01801v2 Announce Type: replace-cross Abstract: Autoregressive video diffusion models enable streaming generation, opening the door to long-form synthesis, video world models, and interactive neural game engines.
arXiv:2510. 09608v2 Announce Type: replace-cross Abstract: Vision-language models (VLMs) could power real-time assistants and autonomous agents, but they face a critical challenge: understanding near-infinite video streams without escalating latency and memory usage.
arXiv:2609.32540v2 Announce Type: replace-cross Abstract: Few-step autoregressive video diffusion generates a long video by splitting the video into temporal chunks and generating chunk-by-chunk, eac...
arXiv:2609.40222v1 Announce Type: new Abstract: When a camera revisits a previously observed region, a video world model should reproduce what was there before. This requires both remembering past ob...
arXiv:2610.35706v1 Announce Type: cross Abstract: Attention over tens of thousands of visual tokens dominates the inference of video diffusion transformers (DiTs). Existing methods reduce this cost t...
An interactive world model must follow the user's actions, remember the places it has shown, and stream in real time. The tension is structural: control wants a short horizon, memory wants an unbounde...
FastBench is a new benchmark that tests streaming Video‑Large‑Language Models (VLMs) on high‑dynamic real‑world video streams, where events occur rapidly and cannot be captured by sparse 1–2 FPS sampling. The benchmark uses a trajectory‑grounded pipeline that generates QA pairs from high‑FPS clips, filters questions answerable at 2 FPS, verifies answers with SAM3 and CoTracker3 trajectories, and incorporates three rounds of human inspection, yielding 306 QA pairs across eight domains and multiple temporal scopes. Experiments show that even the strongest model, Gemini‑3.5‑Flash, achieves only 50.7 % accuracy, and that denser sampling improves performance but saturates when history is compressed; a training‑free baseline, ProactiveFrame, outperforms uniform sampling but still falls short of oracle‑guided focusing.
arXiv:2610.03123v1 Announce Type: cross Abstract: Existing streaming vision-language models (VLMs) continuously perceive and reason over visual streams, but their computational pathways remain fixed...
arXiv:2607. 20125v1 Announce Type: cross Abstract: Autoregressive (AR) video diffusion models have become a promising paradigm for long and streaming video synthesis, but the continuously growing Key-Value (KV) cache makes attention the dominant inference cost, especially at high resolution where each frame contributes many tokens.
arXiv:2608. 07408v1 Announce Type: cross Abstract: We study visual persistence in interactive video world models.