zsxkib / stable-video-face-restoration

SVFR: A Unified Framework for Generalized Video Face Restoration

  • Public
  • 467 runs
  • L40S
  • GitHub
  • Paper
  • License

Input

*file
Preview

Input video file (e.g. MP4).

string

Which restoration tasks to apply.

Default: "face-restoration"

file

An inpainting mask image (white areas will be restored). Only required when tasks includes inpainting.

integer

Number of diffusion steps.

Default: 30

integer

Chunk size for decoding long videos.

Default: 16

integer

Number of overlapping frames between segments.

Default: 3

number

Noise augmentation strength.

Default: 0

number

Minimum guidance scale for restoration.

Default: 2

number

Maximum guidance scale for restoration.

Default: 2

number

Image-to-image noise strength.

Default: 1

integer

Random seed. Leave blank to randomize.

Output

Generated in

This output was created using a different version of the model, zsxkib/stable-video-face-restoration:773f313c.

Run time and cost

This model runs on Nvidia L40S GPU hardware. We don't yet have enough runs of this model to provide performance information.

Readme

SVFR - Video Face Restoration

SVFR (Stable Video Face Restoration) brings old and degraded videos back to life by restoring faces with remarkable detail. This model offers three powerful restoration options:

  1. Face restoration - Enhances facial details and sharpness
  2. Face restoration + colorization - Brings both enhanced details and vibrant colors
  3. Full restoration pipeline - Complete enhancement including masked inpainting

Inputs Required

  • A video file (MP4 or other common formats)
  • Choice of restoration type
  • Optional mask image for inpainting (required only for full pipeline)
  • Random seed for reproducible results

Notes

  • The model works best with clear, front-facing facial footage
  • For inpainting, provide a black & white mask where white areas indicate regions to restore
  • Processing time depends on video length and chosen restoration options

License

Available for non-commercial research purposes only. See SVFR paper for details.

BibTex

@misc{wang2025svfrunifiedframeworkgeneralized,
      title={SVFR: A Unified Framework for Generalized Video Face Restoration}, 
      author={Zhiyao Wang and Xu Chen and Chengming Xu and Junwei Zhu and Xiaobin Hu and Jiangning Zhang and Chengjie Wang and Yuqi Liu and Yiyi Zhou and Rongrong Ji},
      year={2025},
      eprint={2501.01235},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2501.01235}, 
}

Follow me on Twitter/X