The short version: Google Photos, Gemini, ChatGPT's image tools, and Remini's free tier will all repair cracks, tears, and faded faces in old photographs at no cost, but every one of them caps your resolution, adds some invented detail, and works best on a photo that's merely damaged rather than one that's half destroyed. For anything irreplaceable, run it through a free tool first, then decide if it's good enough to stop there.
The free tools that do this job
There are five free routes worth knowing about, and they're not all equal.
- Gemini (Google's AI, free tier in the Gemini app) can take a cracked or faded photo and reconstruct it with a plain-language prompt. You upload the image, describe the damage, and it rebuilds the missing sections.
- ChatGPT's image tools do the same job inside a chat. You upload the photo and ask it to repair tears, remove stains, and sharpen faces. It's free with a standard account, though you're limited to a handful of image generations per day before you hit a wall.
- Google Photos has a built in "unblur" and restore feature buried in its editing tools on mobile, free for anyone with a Google account, and it's quietly good at fixing soft focus and light scratches.
- Remini has a free tier that will do one or two restorations before nudging you toward a subscription. Good for a quick test, not a long term solution if you've got a shoebox of fifty photos.
- GFPGAN, a free open-source face restoration model you can run through Hugging Face in a browser with no account needed, is the one most people have never heard of and it often does a sharper job on faces than the big-name apps.
Pick two of these and run the same photo through both before you commit to one. They don't all interpret damage the same way, and a face that comes out waxy in one tool often comes out convincing in another.
A worked example: the 1958 wedding photo
Say you've inherited a box of your grandparents' photographs and the one you care about, a 1958 wedding photo, has a crease running straight through the bride's face, a water stain across one corner, and the kind of overall fading that's turned the blacks into grey fog.
Here's a realistic process for that exact photo, start to finish:
- Step 1. Scan or photograph it at the highest resolution your phone or scanner allows, minimum 300 dpi if you're using a flatbed scanner. Free AI tools can only work with what you give them. A blurry phone snap of a photo will never restore as well as a real scan.
- Step 2. Upload it to Gemini or ChatGPT with a specific prompt, not a vague one. "Restore this old photo" gets you a mediocre result. "Repair the crease across the bride's face, remove the water stain in the top right corner, restore natural skin tone and contrast without changing anyone's facial features" gets you something closer to usable.
- Step 3. Run the same upload through Google Photos' unblur tool as a second pass, since it handles general softness better than most generative tools do.
- Step 4. Compare both outputs side by side at full zoom, specifically on the faces. This is where free tools fail most often, so check eyes, noses, and mouths closely before you accept the result.
- Step 5. If you want the photo in colour too, that's a separate step. Restoration and colourisation are different jobs even though people lump them together, and the free AI colourising tools work on an already-repaired image, not a damaged one.
That five-step process costs nothing and takes about twenty minutes for one photo. It will not look like a professional restorer's work, but it will look dramatically better than the original, and for most family archives that's the entire point.
What nobody tells you about "free"
Every article on this topic will give you a list of tools and stop there. What most of them skip is this: free AI restoration doesn't repair your photo, it guesses at your photo. When a tool fills in a crease across someone's face, it isn't recovering lost information, it's inventing plausible pixels based on patterns it learned from millions of other faces. Most of the time that guess is close enough that nobody notices. Sometimes it isn't, and you end up with a grandmother whose smile is subtly wrong, or a nose that's a touch too narrow, and you'll never know unless you have another photo of her to compare against.
For photos where accuracy matters, don't trust the first output. Run it twice, compare the two, and if the face changes noticeably between attempts, that's your sign the tool is filling gaps rather than fixing them.
There's also the resolution problem nobody mentions up front. Most free tiers cap your output at somewhere around 1024 to 2048 pixels on the long edge. That's fine on a phone screen. It's not fine if you want to print the photo at 8 by 10 inches for a frame, where you'll start seeing softness the moment you enlarge it. If printing is the end goal, budget for one paid upscale rather than assuming the free version will hold up at size.
And then there's the privacy point that should bother more people than it does. Every free tool that restores your photo does it by uploading your family's faces to a server you don't control. Read the terms before you upload anything sensitive, particularly if the photo includes children or living relatives who haven't agreed to their faces being processed by a third party's AI model.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
When a prompt matters more than the tool
The gap between a mediocre restoration and a good one is usually the prompt, not the app. If you're working in Gemini, the Gemini AI photo editing prompts built for exactly this kind of job will save you the trial and error of figuring out what language gets the best result. The same logic applies inside ChatGPT, where the difference between "fix this photo" and a structured prompt that names the damage, the desired outcome, and what to leave untouched is the difference between a passable edit and a usable one. There's a running list of sixty copy and paste ChatGPT photo editing prompts that covers restoration alongside colourisation, retouching, and background work, which is worth bookmarking if you've got more than one photo to get through.
Family photo restoration has also become one of the fastest-moving trends on social platforms, and the prompts people are using change every few weeks as the models improve. It's worth checking what's currently trending for family photo edits before you start, because a prompt that worked well six months ago often gets outperformed by a newer phrasing once the underlying model updates.
When free stops being good enough
Free tools are the right call for most photos, most of the time. But there's a line where it stops making sense to keep trying for free, and it's worth knowing where that line is before you waste an evening on it.
If the photo is torn into more than two pieces, if a significant portion of a face is missing entirely rather than just faded, or if it's the only surviving image of someone and you can't afford to get it wrong, that's a job for a paid restorer, either a human one or a paid AI service with higher resolution output and more control. Free tools are built for volume and speed, not precision on a single irreplaceable image. Know which category your photo falls into before you start, and you'll save yourself the frustration of expecting miracle results from a tool that was never built to deliver them.
Frequently asked questions
Can AI restore a photo that's torn into pieces?
Free tools can handle a single crease or a small tear reasonably well, but a photo torn into several pieces usually needs the pieces scanned separately and reassembled in editing software first, since most free AI restoration tools assume they're working with one continuous image rather than fragments.
Is it safe to upload old family photos to free AI tools?
Mostly, but check the terms of service first, particularly for photos containing children or living people, since some free tools reserve the right to use uploaded images to train future models unless you opt out.
Why does the restored face look slightly different from the original?
Because AI restoration fills damaged areas by predicting plausible detail rather than recovering lost information, so a badly faded or creased face can come back looking subtly different from the real person, especially around the eyes and mouth.
What's the best free tool for restoring old photos in 2026?
Gemini and ChatGPT's image tools currently give the most control through detailed prompts, Google Photos is the quickest for general softness and fading, and GFPGAN is worth trying specifically for sharpening faces, since it's free with no account required.
