MiniPx
Add to ChromePricingBlogAccount
No. 05 — AI Tools · Cloud HDPro · on our GPU

One old photo, sharpened.

The sharpen slider has been tried, and the print came back gritty rather than clear. That is the slider working as designed: it raises edge contrast without adding detail. To get detail back into a face, something has to rebuild it.

Pro · Cloud HDDeleted on completionUp to 25 MB · 24 MP
Rebuild the detail in an old photoCloud HD · our GPU

Pro to restore your photo →

How it works

  1. Pick the unsharpened original — JPEG, PNG or WebP, up to 25 MB and 24 megapixels.
  2. Our GPU restores the faces and checks each one against your original. If the result drifts too far from the person in your photo, the run is refused rather than handing back an invented face — and you are not charged.
  3. Download the restored PNG at the same resolution, then do any light sharpening last, in your own editor.

This page is the photo restorer, tuned to the difference between sharpening a photograph and rebuilding it.

All Image Tools

AI Tools

Pro Tools

Compress

Convert

Resize

Compress to Size

Compress for Platform

Tools & Utilities

By Device

Compare

Solutions by use case

For Developers & Web

For Sellers & Ecommerce

For Job Applicants & Govt Forms

For Creators & Social

For Work & Documents

Sharpen an Old Photo — Pro, on our server

Sharpening adds contrast, never information. Every sharpen tool ever written — unsharp mask, smart sharpen, the slider in your phone’s editor — works by finding places where light meets dark and pushing the light side lighter and the dark side darker. The eye reads that as crispness. Nothing has been recovered: the same pixels are still there, wearing a heavier outline. On a healthy modern photograph a small amount is genuinely useful, which is why it exists. On an old print it runs into two problems at once. It cannot tell an eyelash from a grain of silver, so it amplifies the grain with equal enthusiasm; and pushed far enough it leaves visible bright and dark halos hugging every strong edge, which is the single most recognisable sign of an over-processed restoration.

Restoration is the other move, and it is not a slider at all. The model here, RestoreFormer++, finds each face in the photograph, reconstructs it at high resolution using what it has learned about the structure of human faces, and blends the result back into your frame. That is why it can return an eye with a defined iris and lashes where the print only had a soft grey oval — it is rebuilding, not accentuating. Two honest limits come with that power. It only works on faces, so the coat, the wallpaper and the garden come back as they were; and because it is reconstructing, every restored face is compared against the same face in your upload, and a run that has drifted from the person in the photograph is refused rather than returned. Nothing is charged for a refused run. The restoration hub describes that check in detail.

Order is where most old-photo projects quietly go wrong. Sharpening first is the classic mistake: you hand the model a picture full of halos and exaggerated grain, and it has no way of knowing those were added five minutes ago rather than captured in 1962. Work the other way round. Start from the least-processed file you can find, restore the faces, and only then resize for wherever the picture is going — a print, a screen, a message. Sharpen last, at final size, and gently. If your problem is grain rather than softness, read fixing grainy old photos first; if it is a scanner that returns everything slightly soft, the scanning workflow page covers the settings that cause it. And if the picture also needs to be larger, enlarge with the on-device upscaler before that final sharpen, not after.

Only one step in that sequence involves an upload, and this is it. Restoration belongs to Cloud HD, the Pro-only, opt-in feature set that runs on our GPU because the model is too heavy for a browser. Your photograph is encrypted in transit, uploaded to our server for that run and deleted when the run finishes, and in all cases within 24 hours; processing happens in the United States on Cloudflare R2 storage and a Modal GPU, and nothing you upload is ever used to train anything. The resizing, the converting, the compression and the sharpening all happen on your own machine, and a general soft-image case with no face in it belongs with unblur image instead.

How it works

  1. Start from the least-processed file: Use the original scan or camera file. Do not sharpen it first — the model reads halos and amplified grain as if they were detail.
  2. Rebuild the faces: Upload here, up to 25 MB and 24 megapixels. Each face is reconstructed and identity-checked against your original.
  3. Do any resizing next: Enlarge or reduce the restored PNG on your own device, because sharpening is always the last thing you do at the final size.
  4. Apply a light sharpen at the end: In your own photo editor, sharpen gently and judge it at 100 per cent. If you can see halos on a collar, you have gone too far.

Frequently asked questions

What is the actual difference between sharpening and restoring?
A sharpen slider increases contrast along edges the image already has, which makes them read as crisper without adding a single new detail. Restoration is a different operation: the model reconstructs a face at high resolution from what it has learned faces look like, then blends it back. One accentuates what survived; the other rebuilds what did not.
Why does sharpening my scan make it look worse?
Because sharpening is indiscriminate. It raises contrast on film grain, scanner noise and JPEG blocking exactly as eagerly as on eyelashes, and past a certain amount it leaves bright and dark halos on either side of every strong edge. On an old print, which is mostly grain and gentle tonal transitions, a heavy sharpen looks gritty and electric rather than detailed.
Does the restored file come back sharper across the whole photograph?
No, and the page should say so plainly. RestoreFormer++ rebuilds the faces it finds and blends them into the frame; the background, clothing and setting return as they were. If a photograph is soft everywhere and has no person in it, this is the wrong tool — the AI photo enhancer lifts the face requirement and works across the whole frame.
Should I sharpen before or after restoring?
Afterwards, if at all. Sharpening first hands the model halos and amplified grain that were never in the original photograph, and it has no way to know those are artefacts rather than detail. Restore from the cleanest copy you have, then apply a light sharpen at the very end, sized for wherever the picture is actually going.
Can I sharpen a photo without uploading it anywhere?
Yes. MiniPx does not ship a sharpen slider — our on-device tools cover cropping, converting, compressing and upscaling — but every desktop and phone photo editor sharpens locally with no upload at all. Use one of those for the final touch, and keep the upload for the one step that genuinely needs a GPU.
Does sharpening happen on my device or on your server?
Neither, on this page: what runs here is restoration, and it runs on our server. The photo is uploaded to our server for that run and deleted when the run finishes, and in all cases within 24 hours, encrypted in transit and never used to train anything. Any sharpening you do afterwards in your own editor never leaves your machine.