---
title: 'NanoVSR: Towards Real-Time Video Super-Resolution on Edge Devices'
emoji: 🔍
colorFrom: yellow
colorTo: green
sdk: gradio
sdk_version: 6.20.0
python_version: '3.12'
app_file: app.py
pinned: false
license: mit
short_description: Real-time 4x video super-resolution on edge devices
---

<div align="center">

# NanoVSR — Real-Time 4× Video Super-Resolution on Edge Devices

[![arXiv](https://img.shields.io/badge/arXiv-2607.10495-b31b1b)](https://arxiv.org/abs/2607.10495) [![GitHub](https://img.shields.io/badge/GitHub-nanovsr-181717?logo=github)](https://github.com/filippawlicki/nanovsr) [![License](https://img.shields.io/badge/License-MIT-green.svg)](https://github.com/filippawlicki/nanovsr/blob/main/LICENSE)

</div>

Interactive demo for **NanoVSR** (ECCV 2026), a fully convolutional, bidirectional-recurrent video
super-resolution model built for edge devices. It reaches **28.64 dB on REDS4** at **27 FPS on a
Jetson Orin NX (25 W)**, with no custom CUDA ops — just plain 3×3 convolutions after
reparameterization.

Upload a low-resolution clip (or try the built-in REDS4 example) and pick one of four model sizes
to see NanoVSR upscale it 4× in real time, with a side-by-side comparison and a before/after slider.
Inference runs on Hugging Face **ZeroGPU**.

## Model zoo

| Model | Params | REDS4 PSNR/SSIM | Orin NX (FPS) |
| :--- | ---: | :---: | ---: |
| NanoVSR-226k | 226k | 28.23 / 0.8057 | 43.86 |
| NanoVSR-644k (baseline) | 644k | 28.64 / 0.8215 | 27.20 |
| NanoVSR-1.7M | 1.7M | 29.15 / 0.8364 | 19.58 |
| NanoVSR-5.4M | 5.4M | 29.73 / 0.8526 | 8.66 |

Checkpoints are downloaded on demand from the [GitHub releases](https://github.com/filippawlicki/nanovsr/releases)
of the main repo and cached for the lifetime of the Space.

## Links

- Paper: [arXiv:2607.10495](https://arxiv.org/abs/2607.10495)
- Code, training and TensorRT deployment: [github.com/filippawlicki/nanovsr](https://github.com/filippawlicki/nanovsr)

## Citation

```bibtex
@misc{pawlicki2026nanovsr,
      title={NanoVSR: Towards Real-Time Video Super-Resolution on Edge Devices},
      author={Filip Pawlicki and Marcel Kańduła and Marcin Pucek and Kamil Dobies},
      year={2026},
      eprint={2607.10495},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2607.10495},
}
```