File size: 5,628 Bytes
791cfc6 d831434 f032d59 1e5d8af f032d59 d831434 f032d59 d831434 f032d59 a5a915f f032d59 a5a915f f032d59 9f6bfa0 f032d59 791cfc6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 | ---
license: apache-2.0
pipeline_tag: video-to-video
library_name: diffusers
---
# [OSDEnhancer] Taming Real-World Space-Time Video Super-Resolution with One-Step Diffusion (arXiv 2026)
**Authors**: [Shuoyan Wei](https://github.com/W-Shuoyan)<sup>1</sup>, [Feng Li](https://lifengcs.github.io/)<sup>2,\*</sup>, Chen Zhou<sup>1</sup>, [Runmin Cong](https://rmcong.github.io)<sup>3</sup>, [Yao Zhao](https://scholar.google.com/citations?user=474TbQYAAAAJ&hl=en&oi=ao)<sup>1</sup>, [Huihui Bai](https://scholar.google.com/citations?user=iXuCUcQAAAAJ&hl=en&oi=ao)<sup>1</sup>
<sup>1</sup>*Beijing Jiaotong University*, <sup>2</sup>*Hefei University of Technology*, <sup>3</sup>*Shandong University*
<small><sup>\*</sup>Corresponding Author</small>
[](https://arxiv.org/abs/2601.20308)
[](https://huggingface.co/W-Shuoyan/OSDEnhancer)
[](https://github.com/W-Shuoyan/OSDEnhancer)
This repository contains the reference code for the paper "[**Taming Real-World Space-Time Video Super-Resolution with One-Step Diffusion**](https://arxiv.org/pdf/2601.20308)".
---
**In this paper, we propose OSDEnhancer, the first framework that achieves real-world STVSR in one-step diffusion.** Given a low-resolution and low-frame-rate video as input, OSDEnhancer generates a high-resolution and high-frame-rate video.
OSDEnhancer begins with a linear initialization to establish essential spatiotemporal structures and adapt the model for one-step reconstruction. It then applies a divide-and-conquer strategy, introducing the temporal coherence (TC) and texture enrichment (TE) LoRAs that progressively specialize in inter-frame dynamics modeling and fine-grained texture recovery, respectively, while collaborating during inference for enhanced overall performance. A bidirectional VAE decoder employs deformable recurrent blocks to leverage the multi-scale structure of the vanilla VAE, enhancing latent-to-pixel reconstruction through joint multi-scale deformable aggregation and inter-frame feature propagation.
## 🔈News
- ✅ **[May 2026]** The inference code and pretrained checkpoints are now available 👉 [](https://github.com/W-Shuoyan/OSDEnhancer) [](https://huggingface.co/W-Shuoyan/OSDEnhancer)
- ✅ **[Jan 2026]** The arXiv version of our paper has been released 👉 [](https://arxiv.org/abs/2601.20308)
## 📚 Installation
```shell
git clone https://github.com/W-Shuoyan/OSDEnhancer.git
cd OSDEnhancer
conda create -n OSDEnhancer python=3.10
conda activate OSDEnhancer
pip install torch==2.8.0+cu128 torchvision==0.23.0+cu128 --index-url https://download.pytorch.org/whl/cu128
pip install -r requirements.txt
```
## 🚀 Usage
### Pretrained Checkpoints
The pretrained checkpoint is available below.
| Model Name | Base Model | Download Link 🔗 |
|---|---|---|
| OSDEnhancer-v1.0 | [CogVideoX1.5-5B](https://huggingface.co/zai-org/CogVideoX1.5-5B) | [🤗 Hugging Face](https://huggingface.co/W-Shuoyan/OSDEnhancer) |
By default, the inference script automatically loads the checkpoint from Hugging Face. For local checkpoint loading, the checkpoint directory should be organized as follows:
```text
ckpt/
├── transformer/
│ ├── config.json
│ ├── diffusion_pytorch_model-00001-of-00002.safetensors
│ ├── diffusion_pytorch_model-00002-of-00002.safetensors
│ └── diffusion_pytorch_model.safetensors.index.json
├── vae/
│ ├── config.json
│ └── diffusion_pytorch_model.safetensors
├── scheduler/
│ └── scheduler_config.json
└── prompt_embeddings/
└── empty.safetensors
```
### Inference
Run OSDEnhancer on an input video:
```bash
python inference.py \
--input demo/input.mp4 \
--output demo/output.mp4 \
--spatial_scale 4 \
--temporal_scale 2
```
For stable inference, we recommend using a GPU with **not less than 80GB of VRAM**. We recommend setting `spatial_scale = 4` and `temporal_scale = 2`. To use a local checkpoint, specify `--ckpt_path`. For long videos or high-resolution inputs, enable chunk-based inference by additionally setting `--chunk_length` and `--overlap`, where `--chunk_length` should satisfy the form of `8N+1`.
## 📧 Contact
If you meet any problems, please feel free to contact us via email: shuoyan.wei@bjtu.edu.cn
## 💡 Cite
If you find this work useful for your research, please consider citing our paper 😊
```shell
@article{wei2026osdenhancer,
title={Taming Real-World Space-Time Video Super-Resolution with One-Step Diffusion},
author={Wei, Shuoyan and Li, Feng and Zhou, Chen and Cong, Runmin and Zhao, Yao and Bai, Huihui},
journal={arXiv preprint arXiv:2601.20308},
year={2026}
}
```
## 📕 License & Acknowledgement
This project is released under the Apache License 2.0. OSDEnhancer is built upon [CogVideoX](https://github.com/zai-org/CogVideo). We also sincerely thank the authors of [DOVE](https://github.com/zhengchen1999/DOVE), [EvEnhancer](https://github.com/W-Shuoyan/EvEnhancer), and [RealBasicVSR](https://github.com/ckkelvinchan/realbasicvsr) for their excellent open-source implementations, which provided valuable references for this project. |