Restore old photos
Blurry faces, sharpened. Old scans, brought back. Free, unlimited, never uploaded.
This tool does not work in Safari. Its WebAssembly engine stops partway through the AI model. Open this page in Chrome, Edge or Firefox — they all work, including on a Mac. On iPhone, install one of those browsers from the App Store: iOS Safari and every iOS browser share the same engine. (340 MB)
Drop old or blurry photos here or click to browse
The AI model downloads once (~340 MB), then everything runs on your device. Photos are never uploaded.
Restoration without handing over your family photos
Every popular photo restorer works the same way: upload your photo to their server, wait, pay credits to download the result. QuietPix runs the same class of neural network (GFPGAN, the research standard for face restoration) directly in your browser. Your grandmother's portrait never leaves your device.
It pairs well with our other tools for old photos: restore the faces here, thencolorize the photo if it is black and white, and upscale it for printing. All free, all local.
Frequently asked questions
How does AI photo restoration work?
A face restoration network (GFPGAN) finds each face in your photo, rebuilds its detail — eyes, skin, hair — and blends the result back in. It runs entirely on your device, so your photos are never uploaded. Only faces are changed; the rest of the photo stays untouched.
Is it really free and unlimited?
Yes. The AI runs on your hardware, so we have no per-photo costs. No credits, no watermark, no signup.
Why is the model download so big?
Face restoration needs a large network (~340 MB) to invent believable detail. It downloads once, is cached by your browser, and afterwards every photo is processed locally. With a modern GPU each face takes a few seconds.
What photos work best?
Photos where a face is at least about 100 pixels tall: old scans, blurry snapshots, low-light phone photos, group pictures. It restores up to 8 faces per photo. If no face is found, the tool says so instead of returning a fake result.
Does it change what the person looks like?
It aims to keep identity while sharpening detail, but it is a prediction, not a recovery of real data. On very blurry faces the result can drift from the real person. Always keep your original.