LayerRecall: layer-wise memory routing for long-horizon consistency in video generation
zju · hf · 2026-08-31
Zhejiang University's LayerRecall is a state-conditioned memory router that selectively injects long-range historical memory into specific video diffusion layers, improving long-video consistency, supervised by cross-horizon prediction matching.
More from Multimodal
- Best settings for training LoRAs on ANIMA models? — XZtext18 · 2026-08-31
- Alibaba Launches Qwen Create: Wan3.0 Video Model & Multi-Agent Teams — 机器之心 · 2026-08-31
- Prompt: Creating a 30-second cinematic horror transformation — LudovicCreator · 2026-08-31
- Discussion on stiffness in AI-generated videos — Toriggle · 2026-08-31
- Midjourney Prompt: A Warm Encounter in the Rainy Forest — tisch_eins · 2026-08-31
- Suno extends Mythos composition into full song versions — repligate · 2026-08-31