Relax Forcing: Relaxed KV-Memory for Consistent Long Video Generation

• URL: http://arxiv.org/abs/2603.21366v1
• Date: Sun, 22 Mar 2026 18:59:24 GMT
• Title: Relax Forcing: Relaxed KV-Memory for Consistent Long Video Generation
• Abstract summary: Autoregressive (AR) video diffusion has recently emerged as a promising paradigm for long video generation.<n>We show that extending generation to minute-scale horizons remains challenging due to progressive temporal degradation.<n>We introduce Relax Forcing, a structured temporal memory mechanism for AR diffusion.
• Score: 73.84423888025171
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Autoregressive (AR) video diffusion has recently emerged as a promising paradigm for long video generation, enabling causal synthesis beyond the limits of bidirectional models. To address training-inference mismatch, a series of self-forcing strategies have been proposed to improve rollout stability by conditioning the model on its own predictions during training. While these approaches substantially mitigate exposure bias, extending generation to minute-scale horizons remains challenging due to progressive temporal degradation. In this work, we show that this limitation is not primarily caused by insufficient memory, but by how temporal memory is utilised during inference. Through empirical analysis, we find that increasing memory does not consistently improve long-horizon generation, and that the temporal placement of historical context significantly influences motion dynamics while leaving visual quality largely unchanged. These findings suggest that temporal memory should not be treated as a homogeneous buffer. Motivated by this insight, we introduce Relax Forcing, a structured temporal memory mechanism for AR diffusion. Instead of attending to the dense generated history, Relax Forcing decomposes temporal context into three functional roles: Sink for global stability, Tail for short-term continuity, and dynamically selected History for structural motion guidance, and selectively incorporates only the most relevant past information. This design mitigates error accumulation during extrapolation while preserving motion evolution. Experiments on VBench-Long demonstrate that Relax Forcing improves motion dynamics and overall temporal consistency while reducing attention overhead. Our results suggest that structured temporal memory is essential for scalable long video generation, complementing existing forcing-based training strategies.

Related papers

• Ring Forcing: Towards Precise Long-Term Memory for Autoregressive Video Diffusion [30.13108213701987]
  Ring Forcing is an autoregressive video diffusion framework designed to robustly construct and precisely utilize long-term memory.<n>Our ring-structured training strategy enforces retrieval from distant history, effectively reconciling the trade-off between strict historical adherence and generative diversity.<n>We present a sparse RoPE mechanism to enable flexible, scalable memory adaptation while fully exploiting pre-trained priors.
  arXiv  Detail & Related papers  (2026-08-27T08:25:32Z)
• DSAR: Dual-Stream Autoregressive Modeling of Temporal Cloth Dynamics for Photorealistic Animatable Avatars [7.198815179389743]
  We introduce a dual-stream autoregressive framework that explicitly models both observable geometric information and implicit internal state.<n> Experiments on challenging datasets demonstrate significant improvements in rendering quality, temporal consistency, and generalization to motion patterns beyond training distributions.
  arXiv  Detail & Related papers  (2026-08-11T05:19:21Z)
• Self Gradient Forcing: Native Long Video Extrapolation [60.00513917489133]
  Self Gradient Forcing restores historical context-gradient gap.<n>Code and models will be released to advance research on autoregressive video generation.
  arXiv  Detail & Related papers  (2026-07-22T16:59:23Z)
• SAGA: Stable Acceleration Guidance for Autoregressive Video Generation [17.967260849031593]
  We propose SAGA, a training-free textbftextitstable textbftextitautoregressive video generation approach.<n>SAGA integrates an acceleration domain spectral guidance objective based on finite-window Slepian projections.<n>Experiments show that SAGA consistently improves temporal quality across multiple autoregressive diffusion models.
  arXiv  Detail & Related papers  (2026-07-09T00:55:10Z)
• DySink: Dynamic Frame Sinks for Autoregressive Long Video Generation [51.58559206569209]
  We propose DySink, a retrieval-based framework that maintains a compact memory bank and selects visually relevant historical frames as dynamic frame sinks.<n> Experiments on minute-long videos show that DySink consistently improves dynamic degree over strong baselines while also achieving higher temporal quality.
  arXiv  Detail & Related papers  (2026-05-20T11:01:01Z)
• A Progressive Training Strategy for Vision-Language Models to Counteract Spatio-Temporal Hallucinations in Embodied Reasoning [49.61652671596548]
  "Multi-image hallucination reasoning", where a massive performance drop between forward and temporal queries reveals a dependence on superficial shortcuts instead of genuine understanding.<n>To mitigate this, we develop a new dataset: Chain-of-temporal building on detailed reasoning into Chain steps and definitive judgments.<n>Our experiments demonstrate that this approach not only improves accuracy but also the forward-backward performance gap from over 70% to only 6.53%.
  arXiv  Detail & Related papers  (2026-04-12T07:48:44Z)
• LoL: Longer than Longer, Scaling Video Generation to Hour [50.945885467651216]
  This work achieves the first demonstration of real-time, streaming, and infinite-length video generation with little quality decay.<n>As an illustration, we generate continuous videos up to 12 hours in length, which, to our knowledge, is among the longest publicly demonstrated results in streaming video generation.
  arXiv  Detail & Related papers  (2026-01-23T17:21:35Z)
• VideoSSM: Autoregressive Long Video Generation with Hybrid State-Space Memory [42.2374676860638]
  Autoregressive (AR) diffusion enables streaming, interactive long-video generation by producing frames causally.<n>Maintaining coherence over minute-scale horizons remains challenging due to accumulated errors, motion drift, and content repetition.<n>We propose VideoSSM, a Long Video Model that unifies AR diffusion with a hybrid state-space memory.
  arXiv  Detail & Related papers  (2025-12-04T07:06:02Z)
• StateSpaceDiffuser: Bringing Long Context to Diffusion World Models [52.92249035412797]
  We introduce StateSpaceDiffuser, where a diffusion model is enabled to perform long-context tasks by integrating features from a state-space model.<n>This design restores long-term memory while preserving the high-fidelity synthesis of diffusion models.
  arXiv  Detail & Related papers  (2025-05-28T11:27:54Z)
• Long-Context State-Space Video World Models [66.28743632951218]
  We propose a novel architecture leveraging state-space models (SSMs) to extend temporal memory without compromising computational efficiency.<n>Central to our design is a block-wise SSM scanning scheme, which strategically trades off spatial consistency for extended temporal memory.<n>Experiments on Memory Maze and Minecraft datasets demonstrate that our approach surpasses baselines in preserving long-range memory.
  arXiv  Detail & Related papers  (2025-05-26T16:12:41Z)
• StPR: Spatiotemporal Preservation and Routing for Exemplar-Free Video Class-Incremental Learning [79.44594332189018]
  Class-Incremental Learning (CIL) seeks to develop models that continuously learn new action categories over time without previously acquired knowledge.<n>Existing approaches either rely on forgetting, raising concerns over memory and privacy, or adapt static image-based methods that neglect temporal modeling.<n>We propose a unified and exemplar-free VCIL framework that explicitly disentangles and preserves information.
  arXiv  Detail & Related papers  (2025-05-20T06:46:51Z)
• Stable Hadamard Memory: Revitalizing Memory-Augmented Agents for Reinforcement Learning [64.93848182403116]
  Current deep-learning memory models struggle in reinforcement learning environments that are partially observable and long-term. We introduce the Stable Hadamard Memory, a novel memory model for reinforcement learning agents. Our approach significantly outperforms state-of-the-art memory-based methods on challenging partially observable benchmarks.
  arXiv  Detail & Related papers  (2024-10-14T03:50:17Z)
• Temporal Attention Unit: Towards Efficient Spatiotemporal Predictive Learning [42.22064610886404]
  We present a general framework of predictive learning, in which the encoder and decoder capture intra-frame features and the middle temporal module catches inter-frame dependencies. To parallelize the temporal module, we propose the Temporal Attention Unit (TAU), which decomposes the temporal attention into intraframe statical attention and inter-frame dynamical attention.
  arXiv  Detail & Related papers  (2022-06-24T07:43:50Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.