No. 08 — AI Tools · Cloud HDPro · on our GPU

Stains, lifted, the paper survives.

Damp, foxing and a tide-mark from a flooded box leave areas rather than lines. Brush the blotches and our GPU rebuilds what the emulsion under them would have shown — an area at a time, honestly reconstructed.

Pro · Cloud HDDeleted on completionUp to 25 MB · 24 MP
Remove stainsCloud HD · our GPU

How it works

  1. Upload the scan — JPEG, PNG or WebP, up to 25 MB and 24 megapixels.
  2. Brush each stain with a small margin. Put every spot in one mask; keep the brush off faces wherever the picture allows.
  3. PowerPaint rebuilds each masked area from what surrounds it and returns a PNG at the scan’s resolution. One HD run of the 10 a month.

This page is the object remover, tuned to area damage on an archive print rather than to a line or an object.

All Image Tools

AI Tools

Solutions by use case

Remove Stains from an Old PhotoPro, on our server

First decide whether anything is actually missing. Damage to a print falls into two categories that look similar on screen and behave completely differently. A tint — the brown of an ageing silver print, the magenta cast of a nineteen-seventies colour photograph, a yellowed varnish — sits over an image that is entirely intact, and the correct fix is a colour adjustment that touches no detail at all. A stain is different: damp, mould, adhesive or a spilled drink has chemically changed the emulsion, and the picture in that area no longer exists to be recovered. Only the second is a mask job. Zoom to 100% and look for surviving structure inside the discolouration; if you can still make out a collar or a doorway through the brown, mask nothing there and spend nothing.

Foxing is the friendly case; a tide-mark is the demanding one. Foxing — the constellation of small rust-coloured spots that damp paper grows — is a scatter of tiny holes each ringed entirely by good picture, and the fill closes them so completely that the print reads as clean. A water stain is the opposite shape: one large, irregular, soft-edged region that can cover a quarter of the frame, and the model must author every pixel of it from the intact border. Over a plain studio backdrop that goes well. Over a patterned wallpaper it goes approximately. Over a group of faces it goes wrong in the particular way that people notice, because a face averaged from other faces is nobody.

Mask generously at the edges and sparingly over people. Stains have soft boundaries where the damage thins out rather than stops, so a mask drawn on the obvious brown leaves a pale halo that the fill blends into. Take a few pixels past the visible edge everywhere. Then do the opposite over faces: pull the mask back to the hairline or the jaw and leave features alone, because there is a model on this site actually trained to rebuild them. Run the cleaned image through face restoration afterwards and it will bring back eyes and mouths with an identity check against your original, which is a far better outcome than an inpainted guess. Line damage — a crease, a scratch, a torn edge — belongs to scratch removal, which is a thin mask and an easier problem, and anything sitting on top of the print rather than in it — a drawing pin, a sticker, a corner mount — is straightforward work for the object remover.

Keep the evidence, and know the terms. Whatever you decide to clean, archive the raw scan untouched and name every derivative so that in twenty years the difference between the document and the restoration is obvious. Colour comes last if the print is black and white, through the colourising tool, and file size for sharing is on-device work in the image compressor, which uploads nothing. Stain removal itself is Cloud HD: Pro-only, opt-in, processed on our GPU in the United States, with the scan and mask encrypted in transit, deleted when the run finishes and in all cases within 24 hours, and never used to train anything. Pro includes 150 runs a month on Monthly and 200 on Annual, of which 10 may be HD runs.

How it works

  1. Scan the stained print: Flatbed at 600 DPI, glass clean, print flat. Crop away the scanner bed on device before uploading anything.
  2. Decide what is loss and what is only tint: Zoom in. Detail still visible inside a discoloured area means a colour correction, not a mask. Blank or eaten emulsion means a mask.
  3. Paint the blotches: Open minipx.com/remove-stains-from-old-photo on Pro, upload the scan and brush each stain with a small margin, keeping masks off faces where possible.
  4. Restore the faces afterwards: Run the cleaned PNG through the face restorer, then add colour if the print is black and white. Keep the original scan beside every derivative.

Frequently asked questions

Which kind of stain comes out best?
Small, round foxing spots on plain background — sky, a wall, a studio backdrop, a tablecloth. Each is a few dozen pixels across with intact picture entirely surrounding it, so the fill has more evidence than it needs. A scattering of forty such spots is still a single mask and a single run, and the print usually comes back looking as though it was never stored badly.
Why does a big water stain over a face come back looking wrong?
Because a face is unique and the model can only average. Across a wall, an invented patch of wall is indistinguishable from a real one. Across a cheek and an eyebrow, the model produces a face-shaped surface that belongs to nobody in particular, and a human viewer detects that instantly. Keep masks off faces where you can, and use the face restorer for the features.
My photo is tinted brown rather than stained. Is that the same job?
No, and treating it as one wastes a run. A tint means the image underneath is intact and only the colour has shifted, which any levels or white-balance adjustment corrects without inventing a pixel. A stain means the emulsion in that area is altered or gone, so there is nothing to correct and something must be rebuilt. Look closely: if detail survives inside the discoloured area, do not mask it.
What about mould, mirroring and that silvery sheen on old prints?
Silvering — the metallic bloom in the dark areas of a print — is a surface effect over image that still exists, so it is usually better handled by re-scanning at a different angle than by masking. Mould growth has eaten the emulsion and is genuine loss, so it is a mask job. Judge by whether you can still see picture through the defect.
Should I remove every last spot?
That is an archival judgement rather than a technical one. A print that has visibly lived is a document of its own history, and stripping it perfectly clean can make an 1890s portrait look like a modern reproduction. Many people clear the damage that distracts from faces and leave the edges and margins alone. Keep the untouched scan either way, so the decision stays reversible.
How is my family photograph handled during the run?
Stain removal is Cloud HD — Pro-only and opt-in, run on our GPU in the United States because the model is far too heavy for a browser. The scan and your mask are encrypted in transit, used once, then deleted when the run finishes and in all cases within 24 hours, and never used to train anything. A refused or failed run costs nothing.