Remove a date stamp from a photo

A date imprint is the one removal on this site where a rigid rule works. Everything else wants judgement: where the object ends, how much of its shadow to take, whether the thing behind it is texture or structure. An imprint is a known quantity — a fixed place in the frame, a fixed colour, a fixed stroke width, and a shape that never overlaps anything the photographer cared about. So this page opens with the corner already searched, defaults to the rectangle rather than the brush, and sends the work to the engine that suits strokes a few pixels wide instead of the one that suits a person on a beach.

The imprint itself is an artefact of how those cameras worked. A small LED behind the film gate exposed the digits directly onto the emulsion between frames, which is why the date is orange rather than white, why it glows into the surrounding grain rather than sitting cleanly on top of it, and why it cannot be separated from the picture the way an overlay drawn by software can. It is not on the photograph. It is in it.

1. 1Open the scan. The bottom quarter of the frame is searched as soon as the picture is decoded. If a run of saturated orange digits sitting on one baseline is there, a rectangle is proposed over it and the tool waits.
2. 2Correct the rectangle, or draw your own. The scan proposes; it does not commit. Draw over anything it missed, or hold E and take back anything it claimed that it should not have. On a software timestamp in white or black it will find nothing at all and say so.
3. 3Grow it by two pixels. An imprint has a halo. The LED exposed the emulsion beyond the stroke itself, the scan softened it further, and the two or three pixels of glow around each digit are what survive an otherwise clean fill as a faint orange ghost.
4. 4Run it. Digit strokes are a few pixels wide, so this goes to the marching fill by default: the boundary advances inward and each pixel is coloured from the intact picture nearest it. On sand, sky, grass or a plain wall the result is usually indistinguishable at any zoom.
5. 5Check it against the original. Drag the divider across the corner. What you are looking for is not the digits — they will be gone — but whether the texture underneath continues, or whether there is a smooth patch where it used to be grainy.
6. 6Save it at the scan's own size. The fill runs again over the full-resolution scan at save time. A 4000-pixel scan is filled at 4000 pixels, not enlarged from the preview, and every EXIF field the scanner wrote is dropped on the way out.

• An imprint over flat sky, sand, grass, a wall or water comes out clean nearly every time, because there is real information a few pixels away in every direction.
• An imprint lying across a strong edge — the line of a roof, a doorframe, the horizon — will leave that edge slightly wrong, because the fill carries a gradient across the gap and not a line.
• The scan reports nothing when it finds nothing. It has not been measured against a corpus of real scans, so its hit rate is an honest unknown rather than a number we can quote at you.
• A timestamp a phone drew in white, a scanner's own label, or a date written on the print in pen are all outside what the scan looks for. The rectangle still takes them out; you just have to draw it.

Why does the scan find some imprints and miss others?

It is a heuristic, not a model, and it is tuned for one specific thing: the orange-to-yellow glow a film compact's imprinting LED left on the negative. It looks in the bottom quarter of the frame for short strokes of that colour, saturated well past anything in the scene, each one filling between a fifth and four fifths of its own bounding box, all sitting on a common baseline. That finds film imprints reliably. It does not find a timestamp a phone or a scanner drew in white, black or grey, because none of those pass the colour test, and it will not find an orange imprint over an orange sunset for the same reason nobody else can see it either.

Can it take the date off a hundred scans at once?

Not today, and the honest reason is that a region is drawn per photograph. The one case where a queue would genuinely help is the same imprint in the same corner across a roll scanned in one session, which is a narrow enough job to be worth building properly later rather than approximately now. A batch eraser that asks you to paint once and applies it blind to a hundred different frames is a worse product than doing them one at a time.

The digits are gone but there is a pale smudge where they were.

Grow the region by two pixels and run it again from the original. A smudge almost always means the region covered the strokes and not the glow around them, so the fill averaged intact picture together with imprint-coloured pixels it thought were intact. If the smudge survives a grown region, the area under the digits had fine texture — pebbles, foliage, knitwear — and the marching fill has flattened it; switching to the patch search is the fix, because it copies real texture in rather than averaging.

Does anything happen to the rest of the scan?

No pixel outside the painted region and its feathered edge is altered, and nothing is re-encoded twice. If you save as PNG the untouched pixels come out bit for bit as they went in. If you save as JPG the whole frame is re-encoded, because JPG gives no other option, which is worth knowing if the scan is already a JPG that has been through several generations.

What about a date printed on the back of the print?

That is a separate photograph of a separate surface, and nothing here will help. Scan the front, keep the back, and write the date into the file name — which is the only place a date on a scan is any use to software anyway, since the EXIF capture fields are stripped here along with everything else.