Fix fading and yellowing, remove dust and scratches, and calm the grain of an old photo, with a brush to add or undo any repair. Faces are never redrawn. Runs entirely in your browser — your photo is never uploaded.
Drop a scan or a phone photo of an old print here, or choose one
JPG, PNG, WebP or HEIC ·
Opening the photo...
Repaired marks are filled in from what is around them; faces are never redrawn. Hold Compare to check.
Restoring the family wedding album?
Share it with everyone in it. Upload an album to AccioPix and every guest receives their own photos on WhatsApp via AI selfie matching — no sorting, no group-chat spam.
Five steps, from the print in the drawer to a clean copy you can share or print. Steps 3 and 4 happen in the tool above, in your browser.
1
Scan or photograph it flat
Scan at 600 DPI (1200 for a small wallet print). No scanner? Lay the print flat in soft daylight and shoot straight down with a phone, filling the frame, with no glare on the surface.
2
Crop to the photo
Crop away the scanner bed or the table. A white or black border left round the print throws off the fading fix.
3
Press Restore
It levels the faded tones and removes the yellow or sepia cast, finds dust and scratches and fills them, and calms the grain. Each step can be switched off.
4
Check it against the original
Hold Compare. Open Review marks to see every repair in red: paint over a crease or tear it missed, erase anything that was really part of the picture.
5
Add color if you like, then save
For a black and white photo, Add color gives a believable AI coloring. Save the full-size JPG, and keep your original scan as well.
Before and after
Public-domain photos from archives, restored with this tool at its default settings unless the caption says otherwise.
Fading and yellowing. A 1920s wedding postcard by Fred C. Palmer, Swindon: tones levelled, sepia cast removed, specks filled. Public domain.
Scratches and dust. A tintype of Captain Dandridge McRae, about 1860, unknown photographer. Public domain.
Creases. A 1914 newspaper portrait. The two wide creases were not found automatically (the halftone dots hide them): they were painted over with the brush and filled with the Better fill. Public domain.
What it repairs, and what it estimates
Our guide to unblurring draws a line: sharpening works on the pixels that are there, restoration has to invent. This tool keeps that line visible.
Repaired from what is there: fading, low contrast and color casts (the tones are stretched and balanced, nothing is added); grain (smoothed, edges kept); softness (a light sharpen, off by default).
Estimated: every pixel under a red mark. A scratch, speck or crease hides what was beneath it, so the fill is the AI's guess from the picture around it. On a thin scratch that guess is almost always invisible; across a wide tear it is a plausible patch, not a recovery.
Estimated, if you ask: color for a black and white photo. A monochrome print never recorded color, so the colors are predicted.
Never done: redrawing faces. Many AI restorers replace a face with a new, sharper one that can look like a different person. This tool has no face model; a face changes only where a mark crosses it.
Scanning tips
The scan matters more than any software. A flatbed at 600 DPI gives a 6 by 4 inch print about 3,600 by 2,400 pixels, enough to print it again at the same size or a little larger (see what DPI means and the photo print size chart). Clean the glass first: dust on the scanner adds new specks. With a phone, turn off the flash, use window light from the side, hold the phone parallel to the print, and take two or three shots to pick the sharpest. Glossy prints in an album sleeve glare; take them out if you can.
Private by design. Family archives stay in the family: the photo is never uploaded, stored or used for training. The AI models (38 MB to find scratches, and 15 MB more to fill a wide crease or tear) download once and run on your own device.
How we tested it
We aged ten of our own wedding photos (faded them, gave them a yellow cast and grain) and drew scratches, fibres, specks and creases on them, so the true damage was known pixel by pixel. On a computer with a graphics chip, about 78% of the pixels it marked were real damage, and it found about 62% of the damaged pixels. What it missed was mostly faint marks on busy texture such as hair and foliage, where they are also hardest to see, and small dark specks: it leaves those alone on purpose, because on a face a dark speck is as often a nostril or a pupil. The brush is there for anything it misses. The fading fix halved the average color error against the original photos. Then we ran it on genuine scans of 1860s to 1930s prints from public archives, which is where the examples above come from.
Frequently asked questions
How can I restore an old photo for free?
Scan or photograph the print, open it in the restoration tool on this page and press Restore. It fixes fading and color cast, fills dust and scratches and reduces grain in your browser, and you save a full-size JPG. It is free, with no sign-up and no watermark.
Can a badly damaged or torn photo be restored?
Partly. Scratches, specks, creases and small tears are filled from the picture around them, and you can paint over a tear or a missing corner for the AI to fill. What it puts there is an estimate: where a face, a hand or writing is missing, nothing can bring back what was really there, so large losses on a face are better left to a professional retoucher.
Does AI photo restoration change the face?
Not here. This tool does not use a face model and never redraws a face. It changes faces only where a scratch or a speck crosses one, and only the few pixels under the mark are filled from their surroundings. Many AI restorers redraw faces, which can make your grandmother look like someone else; hold Compare to check any result against the original.
What resolution should I scan an old photo at?
600 DPI for ordinary prints, 1200 DPI for small ones such as wallet photos or contact prints. A phone photo works too: shoot square on in soft daylight, fill the frame and avoid glare.
Can I restore and colorize the same photo?
Yes. Switch on Add color after restoring a black and white photo. The colors are the AI's best guess: skin, skies and foliage usually look right, but clothes and objects may not match the real ones. The photo colorizer does the same on its own.
Do I need to sign up, and are my photos uploaded?
No sign-up, and nothing is uploaded. The restoration and the AI models run in your browser, on your own phone or computer. The models download once and stay on your device for next time.
When should I use a professional restoration service instead?
When parts of a face are missing, when a photo is torn into pieces, when water or mould has destroyed large areas, or when the print is so valuable that you want a retoucher's judgement on every detail. A free tool handles fading, dust, scratches and small tears well; it cannot know what a lost part of the picture looked like.
Next steps
Add color to a black and white photo with the photo colorizer. Enlarge a small print for framing with Upscale Image (this tool saves at the size of your scan; it does not enlarge). Paint out a stain, a date stamp or a stranger with the object remover, and sharpen a shaky snapshot with Unblur Image.