SwinIR (JPEG) — Restoration AI Model
Rescues an over-compressed JPEG: the blocky squares and mosquito noise go, the picture underneath comes back. Pick the preset — 10, 20, 30 or 40 — that matches how badly it was crushed. Five to fifteen seconds.
What it does
JPEG Artifact Removal takes an over-compressed JPEG and rebuilds the picture underneath it: the blocky squares go, the fuzz around edges goes, and muddy reds and blues come back as colours.
The quality preset tells it how badly the source was crushed - 10, 20, 30 or 40, with 10 the default and the most aggressive. It is a setting you match to the file rather than something the model detects, and the worse the original, the more there is to win back.
When to use it
- Blocky, mosaic-looking squares left behind by compression
- Banding, where a smooth gradient became visible steps
- Edges and fine detail that came out soft or smudged
- Old low-quality JPEGs out of an archive or an old phone
- Images crushed for the web, or a screenshot saved as JPEG
Details
- CategoryRestoration
- Year2021
- LicenseApache-2.0
Compliance & Provenance
- ProviderOpen-source · Specialized
- EU AI Act RiskMinimal Risk
- Art. 50 TransparencyNot applicable
Inputs & Outputs
- ImageInputimage
JPEG-compressed source image with artifacts
- ImageOutputimage
Restored image with reduced compression artifacts
Parameters
- jpeg_qualityintegerdefault 10options: 10, 20, 30, 40
JPEG quality level (10/20/30/40)
Quality preset matching the input JPEG's compression level. 10 (default) — heaviest compression (most artifacts). 20 / 30 / 40 — progressively cleaner inputs. Pick the closest preset; mismatched levels can under-restore or over-smooth.
Tags
Alternatives in Image Restoration
- Image Colorization
Auto-colorize B&W photos (DISCO). 10-30s.
- Image Denoiser
Denoise photos (SwinIR). 5-15s.
- Motion Blur Removal (MSSNet)
Balanced motion-blur removal (MSSNet). 5-15s.
- Real-World Blur Removal
Diffusion-based deblur (highest quality). 1-3 min.