Photo restoration, explained honestly
"Restore your photo in one click" sounds like magic, and most sites are happy to leave it there.
We'd rather you understand what's actually happening: knowing how it works also tells you exactly what it can't do.
You get the real result, not a blurry preview. Free means the full-size restored photo: the same file a paid tool would sell you.
TL;DR
AI restoration models are trained on millions of photos to learn what sharp detail, natural colour and real faces look like, then rebuild a plausible version of your damaged photo from that knowledge. It's convincing synthesis, not recovered original data: great for faded, soft or low-resolution photos, honestly limited on anything physically missing.
This page explains the ai photo restoration tools running behind our free tier, not the category in general. Everything below happens on our own GPU the moment a file finishes uploading: you work in the browser, there is nothing to install, and what you download is simply the finished photo at full resolution, without watermark. The free tier is the whole tool rather than a trial, which is what we mean by free photo restoration. A paid pass does exist, for people with large batches: it raises the number of photos per run, the daily limit and the size ceiling, but it runs the same models and does not restore your photo any better. This is the same photo AI free users and pass holders get.
The same pipeline handles photo repair, picture restoration and plain sharpening, and it helps to know which part does what. The restore photo pass works over the whole image and rebuilds detail from context; the same model that can fix photo blur also softens light scratches, creases and stains. The repair brush is a separate step: it only fills in the area you paint over, using the pixels around it, so it does not find damage by itself. Where the paper of a torn or old damaged photo is physically gone, there is nothing left to rebuild from, so damaged photos with a torn corner come back well up to the edge of the tear and unreliable beyond it. A damage photo with mild fading is a good candidate for AI restoration, because more of the original survives. Faces get their own specialist pass, colour is another separate pass, and each of them is free. Each photo restore starts automatically once the upload finishes, and the image fix stage needs no extra button. It makes no difference whether you upload a JPG or a PNG file, a photo, a scan or a screenshot. We built this AI for photo repair on one simple rule: if you want to restore image detail and add colour, run the passes one after the other and check each result.
The core idea: models that have seen millions of photos
Every restoration model is a neural network trained the same basic way: take millions of clean photographs, artificially degrade copies of them (blur, shrink, add grain, compress, fade), and train the network to turn each degraded copy back into its clean original. Repeat at enormous scale, and the network learns two things: what degradation does to real photos, and what the sharp versions of real-world textures (skin, hair, fabric, brick, foliage) actually look like.
When you then hand it your faded scan, it applies that learned knowledge: "this smudge, in this context, is what a sharp eye usually looks like after fifty years of fading: here is that eye, rebuilt." It doesn't retrieve your photo's lost pixels from anywhere; it synthesises new pixels that are statistically consistent with what remains. That distinction matters, and we'll come back to it.
Upscaling & sharpening: super-resolution
The sharpen & upscale mode uses a super-resolution model, a widely used, well-studied open architecture. Given an image, it outputs a version with 4× the pixels in each direction, filling in the new pixels with learned texture rather than the smooth interpolation a normal photo editor uses.
The difference is easy to see. Classic enlarging (bicubic interpolation) averages neighbouring pixels, so a 4× enlargement is just the same blur, bigger. A super-resolution model instead asks, for every region, "what sharp structure most plausibly produced these blurry pixels?" and draws that. Edges become genuinely crisp, film grain and JPG blockiness get cleaned up in the process, and textures come back with believable detail. It was trained on aggressively degraded images (blur + noise + compression stacked together), which is why it copes so well with old scans, whose damage is always a messy combination rather than one clean defect.
Two practical notes from how we run it: large photos are processed in overlapping tiles so the GPU can handle them, and the input is capped at a sane size: feeding a 60-megapixel scan through a 4× model would otherwise produce absurd files. The better your input, the better the output, which is why a good scan matters: see our scanning guide.
Colourisation: predicting colour from context
The colourise mode does something subtly different. A black & white photo already contains the brightness of every pixel: what's missing is the colour information. The model keeps the brightness channel untouched and predicts only the two colour channels, using context: it has learned from millions of colour photos that skies are blue, foliage is green, skin falls in a certain range, wood is brown. Because the original brightness is preserved, the result keeps every bit of the photo's real structure: the model only "paints within the lines" it's given.
Restoring and colourising are two separate passes: restore first and download the result, then colourise that copy. Both run on the free tier. A photo with strong surviving contrast gets noticeably more convincing colour than a badly faded one. A faded colour print can go through the colouriser as well, but the model ignores whatever photo color is left and predicts new colour from the brightness alone, so do not expect it to bring back the original tints.
Honest limit
Colourisation is a plausible prediction, not a historical record. The model cannot know that your grandmother's dress was actually red: it will pick a colour that's statistically likely for a dress of that shade in that era of photo. Objects without strong colour conventions (cars, clothing, painted walls) are exactly where it guesses most. Treat a colourised photo as an interpretation that makes the past feel closer, not as evidence of what colours things were.
Face restoration: a specialist, not a generalist
Faces get their own mode (restore faces) for a good reason: humans are hyper-sensitive to faces. A generic upscaler that does a beautiful job on a brick wall can leave a face subtly "off" in a way everyone notices instantly. Face-restoration models are trained exclusively on faces, so they carry much stronger prior knowledge about facial structure: how eyes, teeth, and skin texture look at every age and angle. The pipeline detects each face in the photo, restores it with the specialist model, and blends it back into the surrounding image.
These models have a "fidelity" dial: lean one way and the output stays maximally faithful to the degraded input (but restores less); lean the other way and faces come out beautifully sharp (but the model invents more). We run a deliberately faithful setting, because for family photos, resemblance beats beauty.
Honest limit
On a mildly blurred face, the reconstruction is guided by plenty of surviving pixels and stays very true. On a heavily degraded face (a tiny smudge of a face in a group photo), the model has to invent most of the detail, and the result can drift from what the person really looked like. It will look like a real face; whether it's exactly their face is not guaranteed. If identity matters, compare against other photos of the same person.
What none of this can do
- Recover information that was never captured. Text that's illegible in the original stays unknowable: the model may render sharp-looking letters, but they're invented. The same goes for a face that's fully blown out or a torn-away corner. (More on what fading physically destroys in why old photos fade.)
- Guarantee historical accuracy. Everything synthesised is plausible, not proven.
- Fix every photo equally. The sweet spot is photos that are faded, soft, low-resolution or mildly damaged, which, happily, is most old photos. The worst cases improve, but don't become new.
None of this needs a specialist studio. Digital photo restoration of this kind starts from a browser tab: crop if needed, restore, download. Creases, scratches, torn edges and water stains go through the same rebuild-from-context approach, with the limits described above: the model leans on the pixels that survive near the damage and on what it has learned about photos in general. There is no separate paywall for the harder cases.
Try it with the hood open
All of the above runs on our own GPU hardware, which is why the tool is free, with no account and no watermark, and why your photo can be deleted from the machine right after processing. Upload one, drag the before/after slider, and judge the result against exactly the expectations this page has set. There is nothing to download or license, because the models run on our servers. That is also where this differs from photo restoration services that have a person retouch your print and charge per image: this is automatic, free, and limited to what a model can do.
Common questions
What is AI photo restoration?
AI photo restoration uses neural networks trained on millions of photographs to rebuild lost detail in old, faded or blurry images. It synthesises plausible pixels from the context around damaged areas rather than recovering the original data.
How does AI photo restoration work?
A model trained on degraded-to-clean photo pairs learns what sharp detail, natural colour and real faces look like, then applies that knowledge to your photo: it rebuilds edges, fills in plausible texture and reduces compression artefacts automatically.
What types of damage can AI photo restoration fix?
Fading, soft focus, low resolution, heavy compression, scratches and creases all respond well. Physically missing sections, a torn corner or a fully blown-out face, cannot be reconstructed because there is no remaining detail to work from.
Is AI photo restoration free online, or is there a catch?
PicReviver offers AI photo restoration online free: full resolution, no watermark and no card, and old photo restoration works the same on scanned prints as on recent photos. The limits are five photos per run and ten runs a day. A paid pass (€5 for 30 days or €3 for 24 hours) raises those limits and the size ceiling, but it runs the same models, so the result is not better. Some sites offer a free trial that caps resolution or adds a mark; this one doesn't.
What makes a good photo to restore with AI?
Photos that are faded, soft, low-resolution or mildly scratched respond best. The more surviving detail the file holds, the sharper and more accurate the restored version will be: which is why a careful flatbed scan of a print outperforms a quick phone photo of it.
Do AI photo restoration apps really work?
Reasonably well on the cases they're built for. On fading, softness and light damage the models rebuild convincing detail. On a torn-out or missing section they can't, because there is nothing left to learn from. And the detail they add is a well-informed guess, so check faces and text against other photos when accuracy matters.
Is there an AI photo restoration app to download, or software to install?
No. You use the tool in the browser while the restoration runs on our own server, so there is no photo restoration app or software package to install, and it works the same on a Mac, on Windows or on Linux.
Can you repair the physical print?
No. This is an online tool for the digital file: there is no studio behind it and nothing to mail. If you were looking for Chicago photo restoration or another local shop to repair a print by hand, a local conservator is the right call. If it's the digital file that needs fixing, upload it here.
See it work on your own photo
Restore a photo free →
By Sam Ridder, I build and run PicReviver on my own hardware, on my own. Who I am.
ShinobiTools · KvK 42147182 · info@shinobitools.com