What you’ll be able to do

Right, so if you’ve got a box of prints going soft at the edges, or one particular photo of your gran that’s faded to nothing, here’s what’s actually possible. You can scan it properly, fix the creases and spots without turning her into a stranger, sharpen out the blur, sort a yellow cast, and even colourise it if you fancy that. All from your phone or a browser. And you keep the original untouched the whole time, because that bit matters more than people think.

I’ll say this once now and then again later because it’s the whole point: the original scan is the master copy. Everything else is a version. Don’t overwrite it.

Before you start

First thing. Never photograph a photo. Don’t lay it on the table under a lamp and snap it with your phone camera, you’ll get glare, weird angles, a curved edge where the print’s lifted slightly. Use a proper scanning tool or an actual scanner.

Second thing, and this is the one that trips people up: there are two totally different kinds of AI tool here, and mixing them up is how faces go wrong.

Non-generative tools work from what’s actually in the pixels already there, they just clean it up. Denoising, unblurring, removing one specific mark, that’s all “non-generative”. The tool isn’t guessing, it’s tidying.

Generative tools invent new detail. You ask it to “restore this” or “colourise this” and it fills in gaps with stuff that looks plausible but wasn’t necessarily there. Backgrounds, creases, random spots on a wall, fine, let it invent away. Faces are where this goes wrong. A generative tool asked to “fix” a blurry face will happily give you a sharp, symmetrical, slightly wrong face. Same person, technically, but the eyes are a bit off, the jaw’s not quite right, and something in you knows it’s not them even before you can say why.

So: generative for backgrounds and creases and spots. Non-generative for faces. Keep that rule in your head for everything below.

Get set up

You don’t need to buy anything to get started, which is nice.

For the scan itself, PhotoScan by Google Photos (free app, google.com/photos/scan) takes several shots and stitches them together, killing the glare and fixing the perspective automatically. Genuinely good for prints in an album you can’t easily remove. If you can lay the print flat and you’ve got access to an actual flatbed scanner, though, go for that instead, at 600 dpi it’ll beat a phone scan every time.

Once you’ve got your scan, here’s the toolkit, roughly in order of “how much does this touch my privacy”:

Apple Photos, if you’re on an iPhone that supports Apple Intelligence, does Clean Up (removes marks and objects, Apple describe it as handling “even larger objects” now) and Extend (fills in a cropped edge). This one runs on the device. Nothing leaves your phone. I like this a lot for that reason alone.

Google Photos (free) has Magic Eraser for marks and objects, Photo Unblur which Google says “improves out-of-focus blurry photos, both new and old”, and a general AI Enhance. Every edit has a Revert button, which, thank goodness. Worth knowing: some of these are limited on non-Pixel phones or want a Google One subscription, and the “edit by asking” chat feature is US-only for now. Also some tools only live in the app, not on the Google Photos website.

Gemini and ChatGPT can both take an uploaded photo and edit it by prompt. Good for spots, creases, colour casts, that kind of thing. Ask for more than that (like “make this sharper”) and you risk changed faces, same problem as before.

Remini (remini.ai, free app with in-app purchases, also a web app, made by Bending Spoons) is built specifically for this: “enhance, unblur and restore”, repairing old, blurry or scratched photos. It’s good, but it’s generative through and through, it makes faces look sharp by essentially inventing them. Always check the result against the original; the app tells you what the subscription costs before you commit.

Topaz Photo (topazlabs.com/topaz-photo, subscription, check topazlabs.com/pricing) is the serious desktop option. Recover Faces, Upscale, Denoise, Sharpen, dust-and-scratch removal built for archival images. It renders locally on your machine, which is a genuine privacy win. There’s also Topaz for Web, browser-based, with Unblur, Photo Restoration, Face Enhancer, Colorizer and Upscale, though that one does send your photo to their servers.

Adobe Photoshop (adobe.com/uk/products/photoshop.html) has a Remove Tool, Generative Fill, Generative Expand. Free trial available, and there’s a free mobile version with premium features behind a subscription. No free web version, worth knowing before you go looking for one.

Which one you pick depends on how much you trust the internet with your gran’s face. Apple Clean Up and Topaz desktop stay on your device. Everything else is sending the photo somewhere.

Try it yourself

The trick here is one change at a time. Don’t ask a tool to “fix everything”, ask it to look first.

Look at this scan of an old photo. List the damage you can see (creases, spots, fading, blur, colour cast) and tell me which to fix first and which tools to use. Do not edit the image.

That gets you a diagnosis before anything touches the image. Useful, and it costs nothing.

Then, for the actual repair, be specific and set boundaries:

Remove the crease across the middle and the white spots. Reduce the yellow cast. Do not change any face, expression, hair or clothing, and do not add anything. Keep it black and white.

Notice what that prompt does. It names the exact damage, and it explicitly fences off the face. That fence is doing a lot of work.

And once you’ve got a result, before you trust it:

Here is the original scan and the restored version. Compare them and list everything in the restored version that is not in the original: new detail on faces, changed expressions, objects added or removed.

Ask the AI to mark its own homework. It’s not perfect at this, but it catches obvious stuff.

Check the result

Put the original and the result side by side, both at 100% zoom, not fitted to screen. Zoomed out, small changes hide.

Look at the eyes first. Then teeth, if visible. Then the jawline. These are the three places a generative tool most often quietly changes something while “improving” a face.

Check nothing’s been added or removed that wasn’t there. A hand, a bit of background, a second person half out of frame, generative fill sometimes tidies things away that were actually part of the photo.

If there’s any writing in the shot, a date scrawled in the corner, a sign, numbers on a jersey, check it’s identical. Text is a dead giveaway when an AI’s guessed at something.

And check the skin. If it’s gone smooth and slightly waxy, like a shop mannequin, that’s the generative tool doing its “improving” thing. Original skin has texture. Fake-restored skin often doesn’t.

In Google Photos specifically, check the AI info label on the edited photo, it tells you an AI edit happened, which is a handy paper trail if you forget which version is which.

If it doesn’t work

Face looks like a stranger. This is the single most common problem, and it means a generative tool got too involved. Undo it. Try Unblur or Clean Up on a smaller area instead of the whole face, or honestly, just accept a slightly softer photo. Soft-but-recognisable beats sharp-but-wrong every time.

Crease still visible after one pass. Don’t go in for a second aggressive attempt, that tends to make it worse. Several small careful passes work better than one big swing.

Yellow tint everywhere. Fix the colour cast before you run any other AI tool on it, not after. Colour correction first, repair second, always in that order.

Photo’s just hopelessly blurry, like genuinely out of focus at the source. No tool recovers detail that was never captured. Anything you get back is invented, full stop. Worth knowing before you pay for anything.

Photo’s tiny, like a passport-sized print. Don’t upscale it and hope. Rescan at a higher dpi instead, you’ll get more real detail than any upscaler invents.

On privacy, quickly, because it matters: Remini, Topaz for Web, Gemini and ChatGPT all receive your photo on their servers, check their own policies for what happens to it after that. Apple Clean Up and Topaz desktop work entirely on your machine, nothing sent anywhere. And if there’s someone else in the photo, ask them before you post a restored version anywhere public. Someone might not want a photo of themselves, restored or not, circulating without them knowing.

One more thing on colourising, since it comes up: it’s guesswork, always. Nobody actually knows the dress was blue. Either say so when you share a colourised copy, or just keep the black-and-white master alongside it so people can see both.

Questions people ask

Can I restore an old photo for free?

You don't need to buy anything to get started. PhotoScan by Google Photos is a free app for scanning, Google Photos is free, and Remini is a free app with in-app purchases. Topaz Photo is a subscription, and Photoshop has a free trial.

Why does AI restoration make faces look like a stranger?

That's a generative tool getting too involved. Generative tools invent new detail, so a blurry face comes back sharp, symmetrical and slightly wrong. Use generative for backgrounds, creases and spots, and non-generative for faces.

What's the best way to scan an old photo?

Never photograph a photo, you'll get glare, weird angles and curved edges. Use PhotoScan by Google Photos, which stitches several shots and fixes perspective, or a flatbed scanner at 600 dpi, which beats a phone scan every time.

Which photo restoration tools are private?

Apple Clean Up and Topaz desktop work entirely on your machine, nothing sent anywhere. Remini, Topaz for Web, Gemini and ChatGPT all receive your photo on their servers, so check their own policies.

How do I check an AI restored photo is accurate?

Put the original and the result side by side at 100% zoom. Look at the eyes, teeth and jawline, check nothing's been added or removed, check any writing is identical, and check the skin hasn't gone smooth and waxy.

Keep exploring

If you want to go further with editing beyond just repair, there’s a guide here called “How to get AI to edit a photo”. If that box of prints has turned into a filing problem, “How to get AI to sort and organise my files” covers that. And if you’ve got old photos where you genuinely don’t know who or what’s in them, “How to get AI to tell me what is in a photo” is worth a look too.

Sources and review notes

https://www.google.com/photos/scan/ https://www.google.com/photos/editing/ https://support.google.com/photos/answer/6128850?hl=en-GB https://support.google.com/photos/answer/14674995?hl=en-GB https://www.apple.com/uk/apple-intelligence/ https://gemini.google/subscriptions/ https://remini.ai/ https://apps.apple.com/gb/app/remini-ai-photo-enhancer/id1470373330 https://www.topazlabs.com/topaz-photo https://www.topazlabs.com/web/image https://www.topazlabs.com/pricing https://www.adobe.com/uk/products/photoshop.html

Review date: 2026-09-22