Model · ETH Zürich · Image-Enhance
SwinIR
Transformer-based image restoration model. Strong on text, edges, and faces — often produces sharper results than GAN-based upscalers on photographic content.
- Modality
- Image-Enhance
- License
- Apache 2.0 (Open)
- Released
- August 23, 2021
- Last verified
- May 10, 2026
- Runs locally
- Yes
Strengths
- Apache 2.0 — fully open
- Strong on text and fine edges (e.g., scanned documents, screenshots)
- One model handles upscaling, denoising, and JPEG compression artifacts
Weaknesses
- Slower than GAN-based upscalers (Real-ESRGAN) for the same output size
- Smaller community ecosystem than Real-ESRGAN
Try it
| Where | Type | Notes |
|---|---|---|
| GitHub | local | Reference implementation |
| Hugging Face Spaces | hosted | Free demo |
Official sources
Change log
- — Initial entry.
Esc