Restore Old Photos with AI

Every shoebox of family photos tells the same story of slow decay. Prints from the 1950s and 60s have drifted towards a yellow-brown haze as the dyes broke down unevenly. Emulsion has cracked into a fine web where a print was bent, and split outright along the folds where it was carried in a wallet. Corners have gone soft and torn. Silverfish and damp have nibbled at the edges. A fingerprint left in 1974 has etched itself permanently into the surface. And underneath all of that, the lens and the film of the time were never very sharp to begin with, so the faces are softer than memory insists they were.

For decades the only remedy was a restoration studio. A retoucher would scan the print, then spend hours with the clone tool and the healing brush, rebuilding a cheek across a crack by copying skin from beside it, painting in the missing corner by hand, and adjusting curves to pull contrast out of the fog. It was skilled, slow work and it was priced accordingly — often more per photo than the family could justify for anything but the single most treasured portrait.

AI restoration compresses that labour into a few seconds. A generative image model trained on an enormous number of photographs has learned what each kind of damage looks like and, more importantly, what tends to lie underneath it: that a crack across a face is not part of the face, that the fog of a faded print hides a full range of tones, that a soft eye probably had a defined iris and lashes. Given a scan, it repairs the damage and rebuilds the detail while keeping the people in the picture exactly who they were. The result is not a guess pasted over the original; it is the original, healed.

The quality of a restoration is set before the AI ever sees the photo, by the quality of the scan. The model can only work with the information it receives, and a blurry, glare-streaked phone snapshot of a print lying on a carpet gives it very little. Ten minutes spent on capture saves every later step and produces a result that can be printed and framed rather than merely shared.

If you have a flatbed scanner, use it at 600 dpi for prints up to postcard size, and at least 300 dpi for larger ones; clean the glass first, lay the print flat, and save as PNG or high-quality JPG. If you are using a phone, photograph the print in even, indirect daylight — near a window but out of direct sun — laid flat on a neutral surface. Hold the phone parallel to the print, fill the frame, and angle slightly away from the window to avoid glare on glossy paper. Turn off the flash; it creates a hotspot and flattens texture. Take two or three shots and keep the sharpest.

Before restoring, tidy the capture with the on-device tools. Use Crop to remove the table, the border and the album page, and Straighten to square the edges. Do not correct fading or colour yet — the model handles that better when it sees the original tonality, and you can refine afterwards. Larger, sharper input gives the model more to work from; a 12-megapixel capture of a 6-by-4 print is ample, and anything above 50 megapixels may exceed a phone’s memory.

With a clean scan in hand, the restoration itself is a few clicks. The sequence below runs from load to download, including the optional colourisation step and the finishing touches that make the difference between repaired and beautiful.

• Load the scan: drag it into the editor, browse for it, or paste it from the clipboard.
• Open Crop. Cut away the border, the table and any album page, and use Straighten to square the print. Apply crop.
• Open the AI brush and choose Restore. This is a whole-photo operation, so no mask is needed; the hint explains that it repairs scratches, tears, fading and blur.
• Press Apply. The AI works for 10 to 30 seconds and returns the photo with cracks closed, tears filled, fading lifted and detail rebuilt.
• Hold Compare and inspect the faces at 100 per cent: eyes, hairlines, expressions. Identity should be preserved exactly; if anything feels over-smoothed, lower Strength.
• If a specific area still shows damage — a deep tear across a background, a missing corner — paint just that area and use Remove or Replace to finish it.
• Optionally choose Colourise and Apply to add natural colour to a black-and-white print. Restore first, always, so the scratches are not colourised too.
• Open Light: a lift in Contrast and Blacks restores depth to a print that had gone foggy. Use Colour to neutralise a yellow cast, or Black & white in Effects for a clean monochrome.
• Add a touch of Grain in Effects if the original’s texture was part of its charm and the result looks too clean.
• Export as PNG for an archival master or JPG at quality 92 for sharing and printing, at the full resolution of your scan.

The Restore operation targets the whole catalogue of physical and chemical damage. Scratches and cracks in the emulsion are closed with texture that matches the surrounding skin, fabric or wall. Tears and missing corners are filled with plausible continuation of the scene. Dust, specks and the mottling of foxing are cleaned away. Fading — the collapse of contrast into a uniform haze — is reversed so that blacks are black again and the face has shape. Colour shifts from decayed dyes are neutralised. And the general softness of an old lens or an out-of-focus original is countered by rebuilding fine detail: the line of an eyelid, the pattern of a tie, the lettering on a shopfront.

What it does not do is change who is in the picture. The model is instructed to preserve identity, expression, pose and clothing, and in practice a restored portrait is unmistakably the same person — sharper and cleaner, not different. The temptation with generative tools is to “improve” a face, and a responsible restoration resists it: a grandmother’s crooked smile and the particular set of a grandfather’s jaw are the whole point of the photo. If a result ever feels like a stranger, lower Strength so the repair blends with the original, or start from a sharper scan.

Colourisation is the most emotionally striking thing you can do to an old photo, and also the one that most needs to be done in the right order. The model does not know the true colour of your grandmother’s dress; it has learned the statistics of the world — skies are blue, grass is green, skin covers a range of warm tones, brick is red-brown — and it paints those probabilities onto the luminance of the original, keeping every line and grain exactly where it was. The result is a plausible colour version, not a historical record, and it makes a face from another century feel present in a way the grey original does not.

Always restore first. If you colourise a scratched print, the model will faithfully colourise the scratches, and removing coloured cracks afterwards is harder than removing grey ones. Once Restore has done its work, choose Colourise and press Apply. Hold Compare to check that skin tones read as natural and that no colour has bled from a jacket onto a wall behind it. If something is locally wrong — a pink tint on a shirt you know was white — paint just that region and use Replace with a short description like “white shirt”.

Finish with the on-device Colour panel. A colourised photo is often slightly too cool or too intense; a small nudge to Temperature or a pullback on Saturation settles it. If you know a detail — the family car was green, the front door was blue — the Replace operation with a painted mask can put the truth back. Colourise is an AI operation and uses one of your daily twenty, as does Restore, so a full restore-and-colourise takes two.

The AI does the heavy lifting; the Light, Colour, Detail and Effects panels do the finishing, and they run in your browser with no limit. The most common post-restoration adjustment is tonal. Even after the model has lifted the fog, a print from the 1960s often benefits from a deliberate lift in Contrast and a slight pull-down on Blacks to give it weight. Use Highlights to keep a white shirt or a bright sky from glowing, and Shadows to open up a face under a hat brim.

Colour decisions come next. For a colour print with a residual cast, Temperature and Tint can neutralise it — find something that should be neutral, such as a white collar or a grey pavement, and adjust until it reads neutral. For a black-and-white print that the scan has rendered slightly sepia or greenish, the Black & white preset in Effects gives a clean monochrome, and the Intensity slider lets you keep a hint of warmth if you prefer the vintage tone. Faded, Film and Warm presets can recreate a period look on a restored photo that has come back too modern.

Detail is the last stop: a little Clarity adds presence to fabric and hair, Noise reduction smooths residual scanner grain, and a small amount of Grain added back restores the feel of film without restoring the damage.

Most restoration problems trace back to the scan or to over-processing, and both have simple fixes. Before re-running anything, hold Compare and zoom to the faces — that is where it matters and where problems show first.

• Faces look waxy or too smooth: lower Strength so the AI result blends with the original texture, and add a touch of Grain.
• A person looks subtly different: the source was very blurry and the model had little to anchor on. Rescan sharper if you can; otherwise lower Strength and accept a gentler repair.
• A crack or tear is still visible: paint just that line with the AI brush and use Remove. Local passes work at full detail within the painted window.
• The photo is still yellow or foggy: use Light — Contrast up, Blacks down — and Colour to neutralise the cast. The model lifts fading; you set the final tone.
• Glare or a bright patch from the capture remains: that is in the scan, not the print. Rephotograph angled away from the window, or paint the patch and use Remove.
• Colourised colours look wrong or have bled: check whether the scratches were colourised too — if so, Undo, Restore first, then Colourise. Fix local errors with Replace.

Old family photographs are among the most personal images anyone owns, so it is fair to ask what happens to them. Cropping, straightening, light, colour, detail and effects all run in your browser; nothing is uploaded, and there is no limit. When you press Apply on Restore or Colourise, the photo — downscaled to the model’s working size — is sent to the server, forwarded to the image model, returned in ten to thirty seconds, and discarded. Nothing is stored: no image, no mask, no prompt. The button says what it sends before you press it.

The export keeps the resolution of your scan. The AI works at up to 1536 pixels and the result is blended back onto your full-size original, so a 4,000-pixel scan exports as a 4,000-pixel file suitable for printing at the original size or larger. There is no watermark, no reduced-resolution free tier, and no account to create. Save a PNG as the archival master, and a JPG for the family group chat.

Restore and Colourise are AI operations, limited to twenty per day per visitor and shared with the image generator on the same site; the remaining count is shown on the Apply button and resets every twenty-four hours. A restore-and-colourise pass on one print uses two, so a shoebox of forty photos is a project for a few evenings — and a pleasant one. When the allowance is spent, every on-device tool keeps working, so you can crop and straighten tomorrow’s batch tonight.

“Is it really free?” Yes — no account, no watermark, no reduced export. Restore and Colourise are free up to twenty AI operations a day; everything else is unlimited. “Do I need a scanner?” No. A phone photo of the print in even daylight works well; see the capture tips above. “Will the AI change my relatives’ faces?” It is instructed to preserve identity, expression and pose, and in practice faces stay recognisably themselves. Always hold Compare to check, and lower Strength if the result feels over-processed.

“Can it fix a photo that is torn in half?” If you scan both halves together, aligned, it will close the tear; if a piece is missing, it fills it with plausible content. “Can it restore a very blurry photo?” It rebuilds detail consistent with what remains, which can be dramatic — but information that was never captured can only be imagined, so a blurry face comes back plausible rather than exact. “Can I restore a colour photo?” Yes; fading and colour shifts in colour prints are repaired, and Colourise is simply not needed.

The wedding portrait with the crease across it, the school photo gone yellow, the only picture of a great-grandparent, cracked and soft: each is a scan, a click and a few seconds away from looking the way it did the day it was printed. Capture it well, restore before you colourise, finish the tone by hand, and check the faces with Compare. What comes back is the same photograph — the same people, the same moment — with the damage of sixty years lifted off it.

PixGenia’s photo restorer works exactly this way: drop in the scan, choose Restore, and a state-of-the-art image model repairs scratches, tears, fading and blur while keeping the people exactly who they were, blended back into your original at full resolution. Colourise in one more click, finish the light and colour with the unlimited on-device tools, and download — twenty free AI operations a day, no account, no watermark. Start with the photo you have been meaning to save.