About this paper
Disen-Forcing is a distillation-time method for autoregressive video diffusion models that disentangles attention-head specialization to improve long-range consistency and motion dynamics in long-horizon video generation.
Autoregressive video diffusion models (AR-VDMs) enable real-time, long-horizon video generation but degrade rapidly once the inference horizon outruns the training horizon, due to compounding exposure bias. To stabilize long rollouts, recent methods heuristically adopt frame sink, keeping initial frames as a persistent model context, inspired by attention sinks in LLMs. We revisit this design and find that the mechanism by which initial frames stabilize long-range consistency in AR-VDMs is not an LLM-style attention sink; instead, long-range consistency is anchored through a sparse subset of attention heads, while the majority of heads predominantly capture local temporal dynamics. This…
Autoregressive video diffusion modelsAR-VDMslong-horizon video generationinference horizontraining horizonexposure biaslong rolloutsframe sinkattention sinkslong-range consistency
- Upvotes
- 5
- Authors
- 8
- Published
- Nov 2, 2026
Summary and abstract excerpt via Hugging Face Papers ↗ · arXiv.
Trending history
Disen-Forcing: Disentangling Semantic Anchoring from Motion for Autoregressive Video Diffusion was first seen on Hugging Face · Daily papers on Nov 16, 2026. Since then it has appeared in 6 checks on 2 different days. Its best position was #27 of 30, reached on Nov 17, 09:28 UTC. It dropped off the list after Nov 17, 09:28 UTC.