LayerRecall: A State-Conditioned Memory Router for Long-Horizon Consistency in Video Generation

Hugging Face Daily Papers Papers
video-generation memory-routing long-horizon-consistency diffusion-model academic-research

Summary

LayerRecall improves long-video consistency in diffusion models by selectively routing historical memory to specific layers, supervised by cross-horizon prediction matching.

Autoregressive video diffusion enables scalable long-video generation by producing chunks from a bounded recent context. While recency-based caching preserves local continuity, it evicts historical cues needed when subjects, objects, scenes, or attributes reappear. Existing memory mechanisms expose models to nonlocal history, but access alone does not ensure effective use. Our analysis reveals that video DiT layers exhibit distinct preferences for current, recent, and distant context, suggesting that long-range memory requires deciding both what to retrieve and where to use it. We introduce LayerRecall, a current-conditioned, layer-selective memory router that retrieves relevant historical K/V states and injects them only into backbone-specific memory-sensitive layers while preserving local attention elsewhere. To reduce reliance on scarce high-quality long-horizon videos and explicit memory-allocation labels, we further propose Cross-Horizon Prediction Matching (CHPM), which uses a privileged long-context reference to supervise the bounded-memory router in prediction space. Across 100 multi-shot evaluation prompts, LayerRecall achieves the best overall results on MemoBench and MovieBench while matching its backbone on VBench-Long, demonstrating stronger long-range recovery without sacrificing local continuity. Qualitative analyses further reveal memory-guided self-correction, whereby initially mismatched local attributes return to their historical appearance without resetting ongoing motion or scene structure. Additional analyses show cross-backbone portability and negligible inference overhead.
Original Article
View Cached Full Text

Cached at: 09/01/26, 12:10 PM

Paper page - LayerRecall: A State-Conditioned Memory Router for Long-Horizon Consistency in Video Generation

Source: https://huggingface.co/papers/2608.28460 Published on Aug 28

·

Submitted byhttps://huggingface.co/Yixuan-Ding-ZJU

dingon Aug 31

Abstract

LayerRecall selectively routes long-range historical memory into specific video diffusion layers to improve long-video consistency, supervised by cross-horizon prediction matching.

Autoregressive video diffusionenables scalable long-video generation by producing chunks from a bounded recent context. While recency-based caching preserves local continuity, it evicts historical cues needed when subjects, objects, scenes, or attributes reappear. Existing memory mechanisms expose models to nonlocal history, but access alone does not ensure effective use. Our analysis reveals thatvideo DiTlayers exhibit distinct preferences for current, recent, and distant context, suggesting thatlong-range memoryrequires deciding both what to retrieve and where to use it. We introduceLayerRecall, a current-conditioned,layer-selective memory routerthat retrieves relevant historicalK/V statesand injects them only into backbone-specificmemory-sensitive layerswhile preserving local attention elsewhere. To reduce reliance on scarce high-quality long-horizon videos and explicit memory-allocation labels, we further proposeCross-Horizon Prediction Matching(CHPM), which uses a privileged long-context reference to supervise the bounded-memory router in prediction space. Across 100 multi-shot evaluation prompts,LayerRecallachieves the best overall results on MemoBench and MovieBench while matching its backbone on VBench-Long, demonstrating stronger long-range recovery without sacrificing local continuity. Qualitative analyses further reveal memory-guided self-correction, whereby initially mismatched local attributes return to their historical appearance without resetting ongoing motion or scene structure. Additional analyses show cross-backbone portability and negligible inference overhead.

View arXiv pageView PDFProject pageGitHub9Add to collection

Get this paper in your agent:

hf papers read 2608\.28460

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper1

#### Yixuan-Ding-ZJU/LayerRecall Text-to-Video• Updatedabout 5 hours ago • 1

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2608.28460 in a dataset README.md to link it from this page.

Spaces citing this paper1

Collections including this paper1

Similar Articles

Memento: Reconstruct to Remember for Consistent Long Video Generation

Hugging Face Daily Papers

Memento is a subject-reconstruction-guided framework that improves long-form video generation by preserving recurring subjects through memory-based reconstruction and dual-query mechanisms, achieving state-of-the-art performance in long-term subject consistency and cross-shot coherence.

ReWorld: An Interactive World Model with Long-Horizon Memory

Hugging Face Daily Papers

ReWorld presents an interactive world model that decouples short-horizon control from long-horizon memory using mixed attention windows and distribution-matching LoRA distillation, enabling real-time video generation with high action fidelity and recall.