zsxkib / aura-sr

AuraSR: GAN-based Super-Resolution for real-world

  • Public
  • 1.7K runs
  • GitHub
  • Paper
  • License
Run with an API
Playground API Examples README Versions

Examples

View more

Run time and cost

This model costs approximately $0.080 to run on Replicate, or 12 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia A40 (Large) GPU hardware. Predictions typically complete within 111 seconds. The predict time for this model varies significantly based on the inputs.

Readme

✨ AuraSR: GAN Super-Resolution for Images 🖼️

Replicate

AuraSR is a powerful GAN-based super-resolution tool that enhances image clarity and size. Based on the GigaGAN concept, it excels with specific image types.

See AuraSR in action

🎨 Features

  • Upscales PNG, lossless WebP, and high-quality JPEG XL (90+) images
  • Supports scale factors of 2x, 4x, 8x, 16x, and 32x
  • Efficient processing with adjustable batch sizes

⚠️ Important Notes

AuraSR is powerful but has some limitations:

  1. Best results with PNG, lossless WebP, and high-quality JPEG XL (90+)
  2. Sensitive to compression artifacts
  3. No built-in error correction for image imperfections
  4. Ideal for upscaling AI-generated or high-quality uncompressed images

🛠️ Usage

Input Parameters

  • image: The input image to upscale (PNG, WebP, or high-quality JPEG XL)
  • scale_factor: Upscaling factor (2, 4, 8, 16, or 32)
  • max_batch_size: Number of image tiles processed simultaneously (default: 1)

Example

import replicate

output = replicate.run(
    "zsxkib/aura-sr:<VERSION>",
    input={
        "image": open("path/to/your/image.png", "rb"),
        "scale_factor": 4,
        "max_batch_size": 4
    }
)
print(output)

🙌 Acknowledgements

  • fal.ai for the original AuraSR implementation
  • lucidrains for the unofficial PyTorch implementation of GigaGAN

Citation

If you use this model in your research or applications, please cite the original GigaGAN paper:

@article{DBLP:journals/corr/abs-2303-05511,
  author    = {Minguk Kang and
               Jaesik Park and
               Namhyuk Ahn and
               Sungsoo Ahn and
               Kibeom Hong and
               Bohyung Han},
  title     = {GigaGAN: Large-scale GAN for Text-to-Image Synthesis},
  journal   = {CoRR},
  volume    = {abs/2303.05511},
  year      = {2023},
  url       = {https://arxiv.org/abs/2303.05511},
  eprinttype = {arXiv},
  eprint    = {2303.05511},
  timestamp = {Tue, 14 Mar 2023 17:06:10 +0100},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2303-05511.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

License

This model is released under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license.

🐦 Connect

Questions or feedback? Follow me on Twitter @zsakib_ and let’s chat!