Awesome Diffusion Models for Video Super-Resolution
August 3, 2026 · View on GitHub
A curated list of resources for Video Super-Resolution (VSR) using Diffusion Models.
一个关于 基于扩散模型的视频超分辨率 (VSR) 精选资源的列表。
⚡️ Note: This list is updated daily by following arXiv. If you find anything missing, feel free to open an issue! 😊
📖 Table of Contents / 目录
🌟 Introduction / 简介
This repository focuses on collecting resources related to Video Super-Resolution (VSR) using Diffusion Models, including papers, open-source code and datasets. It serves as a one-stop reference for researchers and developers interested in this field.
本仓库专注于收集 基于扩散模型 (Diffusion Models) 的 视频超分辨率 (VSR) 相关资源，包括论文、开源代码、数据集，为对该领域感兴趣的研究者和开发者提供一站式参考资料。
📄 Papers / 论文
The following papers are arranged in chronological order based on their posting time on arXiv.
以下的论文排列顺序是按照在 arXiv 上发布的时间顺序排列的。
TODO: VSR models inherent in video foundation models.
2026
| Title | Published | Code | Keywords |
|---|---|---|---|
| OSDEnhancer: Taming Real-World Space-Time Video Super-Resolution with One-Step Diffusion | Arxiv2026(Jan) | Demo | T2V(CogVideoX1.5-5B)-based, one-step |
| LSGQuant: Layer-Sensitivity Guided Quantization for One-Step Diffusion Real-World Video Super-Resolution | Arxiv2026(Feb) | Github | T2V(Wan2.1-1.3B)-based, quantization for one-step VSR |
| Tiled Prompts: Overcoming Prompt Underspecification in Image and Video Super-Resolution | Arxiv2026(Feb) | I2V(I2VGen-XL)-based, Use Qwen3-VL-8B-Instruct for stronger prompt | |
| D^2-VR: Degradation-Robust and Distilled Video Restoration with Synergistic Optimization Strategy | Arxiv2026(Feb) | T2I(SD x4 Upscaler)-based | |
| Improved Adversarial Diffusion Compression for Real-World Video Super-Resolution | ICLR2026(Feb) | T2I(SD2.1)-based, one-step | |
| SparkVSR: Interactive Video Super-Resolution via Sparse Keyframe Propagation | ECCV2026(Mar) | Github | I2V(CogVideoX1.5-5B-I2V)-based, one-step |
| DUO-VSR: Dual-Stream Distillation for One-Step Video Super-Resolution | CVPR2026(Mar) | Github | T2V(Wan2.1-1.3B)-based, one-step |
| InstaVSR: Taming Diffusion for Efficient and Temporally Consistent Video Super-Resolution | Arxiv2026(Mar) | T2I(SD2.1)-based, one-step | |
| NTIRE2026 Challenge on Short-form UGC Video Restoration in the Wild with Generative Models: Datasets, Methods and Results | NTIRE2026(Apr) | UGC video, mainly T2V-based and one-step | |
| DiffST: Spatiotemporal-Aware Diffusion for Real-World Space-Time Video Super-Resolution | Arxiv(May) | Github | T2V(Wan2.1-1.3B)-based, one-step |
| DTG-Restore: Training-Free Diffusion Refinement for Generative Video Super-Resolution | Arxiv(Jun) | T2V(Wan)-based, training-free | |
| LiteVSR: Lightweight Adaptation of Frozen Diffusion Transformers for Video Super-Resolution | Arxiv(Jun) | T2V(Wan2.2-TI2V-5B)-based | |
| SwiftVR: Real-Time One-Step Generative Video Restoration | Arxiv(Jun) | Github | T2V(Wan2.2-TI2V-5B)-based, one-step |
| RGFVR: Reference-Guided Face Video Restoration with Flow Matching | Arxiv(Jun) | T2V(Wan2.1-1.3B)-based, face restoration | |
| SATB-VR: Training Few-Step Video Restoration Diffusion Model using SNR-Aware Trajectory Blending | Arxiv(Jun) | Github | T2V(CogVideoX1.5-5B)-based |
| DTI: Dynamic Trajectory Initialization for Generative Face Video Super-Resolution | ECCV2026(Jun) | Github | T2V(Wan2.1-1.3B)-based, face restoration |
| AVSR-Diff: Scale-Agnostic Diffusion Priors for Temporally Consistent Arbitrary-Scale Video Super-Resolution | ECCV2026(Jul) | Github | T2I(SD x4 Upscaler)-based |
| FADRA: Frequency-Aware Diffusion with Residual Adaptation for Video Face Restoration | Arxiv2026(Jul) | T2V(Wan2.1-1.3B)-based, face restoration | |
| LPM: Industrial-Scale Generative Video Restoration | Arxiv2026(Jul) | Train from scratch | |
| TRaM-VSR: Importance-Aware Token Routing and Merging for One-Step Diffusion Video Super-Resolution | Arxiv2026(Jul) | Github | Plug-and-play module |
2025
| Title | Published | Code | Keywords |
|---|---|---|---|
| SeedVR: Seeding Infinity in Diffusion Transformer Towards Generic Video Restoration | CVPR2025(Jan) | GitHub | Train the full model on 256 NVIDIA H100-80G GPUs |
| DiffVSR: Enhancing Real-World Video Super-Resolution with Diffusion Models for Advanced Visual Quality and Temporal Consistency | ICCV2025(Jan) | GitHub | T2I(SD x4 Upscaler)-based |
| STAR: Spatial-Temporal Augmentation with Text-to-Video Models for Real-World Video Super-Resolution | ICCV2025(Feb) | GitHub | T2V(CogVideoX-5B)-based |
| UltraVSR: Achieving Ultra-Realistic Video Super-Resolution with Efficient One-Step Diffusion Space | ACMMM2025(May) | GitHub | T2I(SD)-based, one-step |
| DOVE: Efficient One-Step Diffusion Model for Real-World Video Super-Resolution | NIPS2025(May) | GitHub | T2V(CogVideoX1.5-5B)-based, one-step |
| LiftVSR: Lifting Image Diffusion to Video Super-Resolution via Hybrid Temporal Modeling with Only 4×RTX4090s | Arxiv2025(Jun) | GitHub | T2I(PixArt-α)-based |
| DLoRAL: One-Step Diffusion for Detail-Rich and Temporally Consistent Video Super-Resolution | NIPS2025(Jun) | GitHub | T2I(SD2.1)-based, one-step |
| SeedVR2: One-Step Video Restoration via Diffusion Adversarial Post-Training | ICLR2026(Jun) | GitHub | Based on SeedVR, one-step |
| SimpleGVR: A Simple Baseline for Latent-Cascaded Video Super-Resolution | Arxiv2025(Jun) | T2V-intialized, train on 16 NVIDIA H800-80G GPUs | |
| DAM-VSR: Disentanglement of Appearance and Motion for Video Super-Resolution | SIGGRAPH2025(Jul) | GitHub | I2V(Stable Video Diffusion)-based |
| TurboVSR: Fantastic Video Upscalers and Where to Find Them | ICCV2025(Jul) | I2V(LTX-Video)-based | |
| RealisVSR: Detail-enhanced Diffusion for Real-World 4K Video Super-Resolution | Arxiv2025(Jul) | Github | T2V(Wan2.1-1.3B)-based |
| Semantic and Temporal Integration in Latent Diffusion Space for High-Fidelity Video Super-Resolution | Arxiv2025(Aug) | T2I(SD2.1)-based | |
| Vivid-VR: Distilling Concepts from Text-to-Video Diffusion Transformer for Photorealistic Video Restoration | ICLR2026(Aug) | Github | T2V(CogVideoX1.5-5B)-based |
| OS-DiffVSR: Towards One-step Latent Diffusion Model for High-detailed Real-world Video Super-Resolution | Arxiv2025(Sep) | T2I(SD2.1)-based, one-step | |
| Towards Redundancy Reduction in Diffusion Models for Efficient Video Super-Resolution | Arxiv2025(Sep) | Github | T2V(Wan2.1-1.3B)-based, one-step |
| Asymmetric VAE for One-Step Video Super-Resolution Acceleration | Arxiv2025(Sep) | Github | T2V(CogVideoX1.5-5B)-based, one-step |
| PatchVSR: Breaking Video Diffusion Resolution Limits with Patch-wise Video Super-Resolution | CVPR2025(Sep) | Project | T2V-based |
| L∀TINO: Latent Video Consistency Inverse Solver for High Definition Video Restoration | Arxiv2025(Oct) | ||
| InfVSR: Breaking Length Limits of Generic Video Super-Resolution | ICML2026(Oct) | Github | T2V(Wan2.1-1.3B)-based, one-step |
| UniMMVSR: A Unified Multi-Modal Framework for Cascaded Video Super-Resolution | Arxiv2025(Oct) | GitHub | T2V-based, hybrid-modal conditions |
| MoA-VR: A Mixture-of-Agents System Towards All-in-One Video Restoration | JSTSP2025(Oct) | GitHub | Agent-based video restoration |
| FlashVSR: Towards Real-time Diffusion-based Streaming Video Super-Resolution | CVPR2026(Oct) | GitHub | T2V(Wan 2.1–1.3B)-based, one-step |
| Rethinking Diffusion Model-Based Video Super-Resolution: Leveraging Dense Guidance from Aligned Features | CVPR2026(Nov) | Github | T2I(SD x4 Upscaler)-based |
| STCDiT: Spatio-Temporally Consistent Diffusion Transformer for High-Quality Video Super-Resolution | CVPR2026(Nov) | GitHub | T2V(Wan2.1 T2V-1.3B and Wan 2.1 I2V-14B)-based |
| CreativeVR: Diffusion-Prior-Guided Approach for Structure and Motion Restoration in Generative and Real Videos | Arxiv2025(Dec) | Project | T2V(Wan2.1-1.3B)-based |
| Stream-DiffVSR: Low-Latency Streamable Video Super-Resolution via Auto-Regressive Diffusion | Arxi2025(Dec) | GitHub | T2I(SD x4 Upscaler)-based |
2024
| Title | Published | Code | Keywords |
|---|---|---|---|
| Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution | CVPR2024(Dec) | GitHub | T2I(SD x4 Upscaler)-based |
| MGLD-VSR: Motion-Guided Latent Diffusion for Temporally Consistent Real-world Video Super-resolution | ECCV2024(Dec) | GitHub | T2I(SD2.1)-based |
📊 Datasets / 数据集
Train
- LSDIR - High-quality image dataset that contains 84,991 training images, 1,000 validation images, and 1,000 test images.
- REDS - 300 video sequences with resolution of 720×1280, and each video has 100 frames.
- OpenVid-1M - A high-quality text-to-video dataset, and all videos in the OpenVid-1M dataset have resolutions of at least 512×512.
- WebVid-2M - A large-scale dataset containing 2.5M text-video pairs with resolution of 336×596.
- YouHQ-Train - 38,576 videos witwh resolution of 1080 × 1920 for training, each of which has around 32 frames.
- UltraVideo - A high-quality test-to-video dataset, containing 42K short videos (3∼10 seconds) and 17K long videos (≥10 seconds) with 4k/8k resolution.
- SpatialVID - A large-scale video dataset containing 2.7M 1280x720 clips. A subset of it SpatialVID-HQ owns 365k clips with higher criterion on aesthetics and motion quality.
- Sekai - A large-scale long video dataset containing Sekai-Real and Sekai-Game datasets, which provides 6620 hours of 720p real-world videos and 60 hours of 720p game-synthesized videos, respectively.
Test
- RealVSR - The dataset consists of 500 LR-HR sequence pairs with real degradation, each of which has 50 frames in length and 1024×512 pixels in size.
- YouHQ-Test - 40 video clips for evaluation, each of which has around 32 frames.
- RealisVideo-4K-Test - 1,000 high-definition video(4K)-text pairs.
- SPMCS - 30 different videos, each of them contains 31 frames.
- UDM10 - 10 video sequences, each containing 32 consecutive frames with a resolution of 720x1272.
- VideoLQ - 50 video sequences, with each sequence containing a variable number of frames ranging from 33 to 100.
:star: If it is helpful to your videos or projects, please help star this repo. Thanks! :hugs: