Upload a scan of a faded, torn or yellowed print and get back a version with the contrast, edges and color recovered, not invented. Picasso IA restores creases, scratches and soft focus, and treats colorizing black-and-white photos as a separate, optional step.
A photograph that survived a house move, a flood or fifty years folded inside a wallet is not raw material, it is evidence. Old photo restoration takes a scan of a faded, cracked or yellowed print and recovers the contrast, color and edges the print already has, rather than a stranger's guess at what used to be there. Done honestly, it gets a face legible again without inventing a person who was never in the frame.
Every restoration model works from what you give it, not from the physical print in a drawer. A photo taken at an angle, under a light that throws glare across a face, or at low resolution, hands the model less to work with than the print actually contains, and the result comes back softer than it should, because detail never captured cannot be recovered later.
The fix is a few extra minutes. Lay the print flat, light it evenly from two sides so no glare spot sits over a face, and fill the frame without cropping. A flatbed scanner at 600 dpi or higher is the gold standard, but a phone photo in daylight with the print flat on a table works for most family snapshots. If faces still look blurred after a first attempt, a more careful second scan is usually the fix, not a different model.
Restoration targets damage that changed how a photo looks without erasing what was recorded: fading that washed out contrast, a yellow or orange color cast from decades of aging, creases and scratches across the emulsion, water stains that bloom across a corner, and general softness from a low-quality print. In each case the scene is still there in the pixels, just degraded, and the model's job is to recover it.
What it cannot honestly do is invent information the print never held. If a face is a featureless smear because the negative was overexposed there, no model can tell you what that person looked like, it can only place a plausible guess where the smear was. That guess might resemble them or it might not, and nothing in the photo lets you tell which. Restoration recovers degraded detail; it does not resurrect detail that was never captured.
Picasso IA's restoration and upscaling models live in the Toolkit alongside the rest of the 488-model catalog, so the same account can run a dedicated upscale pass afterward if the result still needs resolution for printing.
Restoring and colorizing get bundled together, but they are different operations with different honesty standards. Restoring recovers what the photo already recorded, contrast, edges, tone. Colorizing adds information the black-and-white print never captured at all, because a monochrome photo never recorded what color anyone's coat or eyes actually were.
That makes colorizing a judgment call, not a correction. A colorization model infers plausible colors from context, skin tone from shading, common fabric colors for the era, foliage green from shape, and those inferences can be wrong in ways nobody can verify from the photo alone. Grandpa's jacket might come back navy when it was brown, and there is no way to check unless someone who remembers the day is around to say so.
Treat colorizing as an artistic interpretation you are choosing to add, and keep an uncolorized restored version alongside it. Anyone researching genealogy or documenting a photo for the record generally wants the restored black-and-white version treated as authoritative, with the colorized copy labeled as an interpretation.
Pro tip: run restoration and colorization as two separate passes. Restore first, check the result against the original for anything invented rather than recovered, and only then decide whether to colorize. Separating the steps makes it obvious which changes came from recovering real detail and which came from a color guess.
Not every kind of damage responds the same way, and knowing what to expect saves a round of disappointment. The table below reflects how each common type generally behaves.
| Damage type | What restoration typically recovers | What to expect |
|---|---|---|
| Fading and yellowing | Contrast and a more neutral color balance across the image | Usually the strongest, most reliable result |
| Creases and tears | The line closes, tone and texture fill in from surrounding areas | Works well unless the crease crosses a face |
| Water stains and blooms | The stain lightens and blends into surrounding tone | Good outside faces, riskier over key features |
| Heavy scratching | Scratch lines disappear from flat areas like sky or backgrounds | Reliable on backgrounds, harder across hair or fine detail |
| Extreme blur or soft focus | Some sharpening, within the limits of what the scan captured | Modest gains; a badly out-of-focus original stays soft |
Two patterns repeat across every row. Damage in flat, low-detail areas restores more convincingly than damage crossing a face, because less information is genuinely lost there. And a photo with several problems stacked together restores best if you address the worst issue first and re-scan, rather than expecting one pass to fix everything.
An honest restoration tool tells you where it stops working, because the alternative is a result that looks convincing but is quietly wrong.
None of these limits make restoration less useful, they define the honest promise: recovering what a damaged photo still holds, not manufacturing a memory it never had.
The process is short enough to run on a whole shoebox of prints in an afternoon, and the real cost is the time spent getting a clean scan of each one first.
Checking a run's cost takes ten seconds on the pricing page, since restoration and upscaling draw from the same credit pool as every other model.
Yes, and fading is the damage type restoration handles most reliably, because the image content is still present, only its contrast and color balance have drifted. The model rebalances tone across the frame rather than reconstructing anything lost. A print that looks almost gray from decades of light exposure typically comes back with faces and clothing clearly separated again, since that structure was compressed, not destroyed.
Restoring recovers information the photo already recorded but damage or fading obscured, contrast, edges, sharpness. Colorizing adds color to a photo that never recorded color at all, so every color in the result is an informed guess. Treat the two as separate decisions: restore first to recover what is genuinely there, then decide separately whether to add an interpreted color version.
Only partially. If enough shading and edge detail survive to suggest a face's structure, restoration can sharpen and rebalance what is there. If the damage erased the face entirely, any output is a generic plausible face rather than a recovery of the actual person, and should not be treated as an accurate likeness. Restoration recovers detail that is degraded, it cannot resurrect detail that is gone.
Not verifiably. A colorization model infers likely colors from context, skin tone from shading, common fabric colors for the era, but a black-and-white photo never recorded what color anything actually was. The result can look convincing and still be wrong. Treat colorized versions as an interpretation you chose to add, not a factual record, and keep the restored black-and-white version as the reference copy.
A phone photo works for most family snapshots, taken carefully: print flat on a table, even light from two sides with no glare, camera as close as possible while still capturing the whole photo. A flatbed scanner at 600 dpi or higher gives a better starting point, especially for photos you plan to print again. Either way, how much genuine detail the capture contains is the single biggest factor in the outcome.
Restoration and upscaling models live in the Toolkit alongside the rest of the catalog. The right choice depends on what the photo needs, upscaling for softness and low resolution, or a restoration-focused model for creases and fading. With 488 models available, there is no single fixed answer; running the same scan through two options is the fastest way to find what handles a particular photo well.
Yes, fixing both in one generation is usually fine for moderate damage. For a print with several serious problems stacked together, results tend to improve if you address the worst issue first, evaluate, then run a second pass rather than expecting one prompt to solve everything. Re-scanning after the crease has been flattened also helps more than most people expect.
No. The digital restoration is a copy that recovers detail from the print, not a replacement for it. Physical prints keep degrading with time regardless of digital restoration, and the original remains the only true source if a restoration ever needs redoing or rescanning at higher resolution later. Keep it stored flat, dry and away from direct light, even once you have a version you are happy with.
Yes. Once you have a restored still, an image-to-video model can animate it with a gentle camera push or subtle parallax, a common way to bring a family photo into a slideshow or memorial video. Keep the motion slow, since old photos already carry some softness, and aggressive motion tends to expose that rather than hide it. A restored, sharpened frame gives the video model more to work with than the original scan would.
Less than fading or a crease. Sharpening can meaningfully improve a slightly soft photo, but one that was badly out of focus when taken, or scanned at very low resolution, has less genuine information to work with, and no model can recover detail the capture never recorded. The honest outcome is a modest improvement, and re-scanning at higher resolution, if the print still exists, usually helps more than extra passes on the same file.
That box of prints in the drawer is not getting less fragile while it waits. Scan the first one and upload it to the Toolkit to see what restoration recovers before deciding whether colorizing it is worth doing too.
