Restore an Old or Blurry Photo
AI redraws faces, removes noise and softness, and can upscale the resolution. Processing runs locally on our own hardware, your photos never go to the cloud.
Before and after: an old photo restored
Drag the slider to compare the original with the restored version. Damage, blur and noise removed by AI. A real result from this tool.
When restoration helps
- Old archival shots with scuffs and noise
- Blurry portraits where the face turned to mush
- Photos from cheap cameras or poor scans
- Family archive photos you need to enlarge and print
- Old JPEGs with visible compression artifacts
What the AI does
It finds faces in the photo and redraws them while keeping the person's features. At the same time it lifts the overall quality of the frame, removing softness, noise and artifacts. You can also upscale the resolution 2x or 4x.
Modes
Restore only (×1)
Resolution stays the same, only sharpness improves and faces are redrawn. Perfect if the source is already high resolution.
Restore + ×2
Also doubles the frame. Good for archival shots of 1000-2000 pixels on the long side.
Restore + ×4
Maximum upscale, for very small sources. Do not use on photos larger than 1500 pixels: the file will be huge and processing slow.
How to use it
- Drag a photo into the upload area
- Pick a scale (×2 by default)
- Click Restore
- In 3-30 seconds get a JPG
FAQ
Will the AI change the faces?
The model tries to keep the person's features: the set of the eyes, the shape of the nose, the oval of the face. But it is a reconstruction, not the real history of the shot. On heavily blurred photos it may "add" things: the model invents what it cannot see. For official documents AI restoration is not suitable.
What about profile faces and groups?
The detector finds several faces and processes each one separately. Profile faces come out worse, only partly redrawn. Small faces in a crowd may be skipped.
Does it add a face where there is none?
No. If the detector finds no faces, the AI just runs the background through the upscale and returns the result.
Any limits?
10 photos per minute, 60 per day per IP. Up to 25 MB, up to 3000 pixels on the long side on input. If larger, it is downscaled automatically.
What about badly damaged photos?
Deep scratches, torn-off corners and missing fragments are not something the AI will paint in. That is already manual layer-by-layer retouching in Photoshop.
Related tools
- Enlarge photo, if you only need an upscale without working on faces
- Remove background, AI cutout
- HEIC to JPG, converting iPhone photos
- Colorize black and white photos, turn a black and white photo to color
Family archive, torn-off corners, serious damage, leave a request, and we will retouch it by hand in Photoshop.
How to get the best result
- Crop everything that is not the photo itself: the white edge of the print, the mat, the table, the shadow of your phone. The model processes the whole frame, and the margins throw off tone and colour on the image itself.
- A scan beats a photo. If you have no scanner, shoot the print straight from above in even daylight, without flash or glare on the glossy surface.
- Upload the maximum size: scan at 300-600 dpi. From a small file the model invents detail instead of recovering it.
- Scratches, stains, grain and fading are fixed well. Torn-off parts, lost faces and handwriting the model will not rebuild; that is a job for a retoucher.
- Restore before colourising and upscaling: a clean, contrasty source gives a better result at every following step.
If a face looks unlike the person after restoration, crop tighter around the face and try again: the model gets more detail to work with.