base_model:
- Wan-AI/Wan2.1-T2V-1.3B
language:
- en
license: apache-2.0
pipeline_tag: text-to-video
DecMem: Towards Minute-Long Consistent World Generation with Decoupled Memory
DecMem is a decoupled memory architecture designed for consistent, long-horizon world generation. It employs Sparse Global Memory for efficient fine-grained access to global history and Anchored Local Memory for stable and high-quality extrapolation. This approach enables minute-level controllable long video generation with high fidelity and consistency.
Project Page | Paper | Code
Checkpoints
Download the Wan2.1 backbone (VAE + tokenizer weights used by the pipeline):
huggingface-cli download Wan-AI/Wan2.1-T2V-1.3B \
--local-dir-use-symlinks False \
--local-dir wan_models/Wan2.1-T2V-1.3B
Download DecMem trained checkpoints:
huggingface-cli download KlingTeam/DecMem --local-dir checkpoints
Checkpoint layout expected by training / inference scripts:
checkpoints/
└── decmem.pt # released weights
Quick start
We provide example video-pose pairs for quick inference. The inference is performed in a block-by-block causal denoising manner with KV cache.
To run the inference, follow the installation instructions in the official repository and run:
bash scripts/infer_example.sh
Citation
If you find our work helpful, please cite our paper:
@misc{yang2026decmemminutelongconsistentworld,
title={DecMem: Towards Minute-Long Consistent World Generation with Decoupled Memory},
author={Zhenhao Yang and Xiaoshi Wu and Zhengyao Lv and Xiaoyu Shi and Xintao Wang and Pengfei Wan and Kun Gai and Kwan-Yee K. Wong},
year={2026},
eprint={2605.31336},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2605.31336},
}