What fading actually is
A photographic print holds its picture in dye or in metallic silver, and both of them change over decades of light, heat and whatever was in the air. The picture does not blur and it does not move. What happens is narrower than that: the darkest part of the print stops being as dark as it was and the lightest part stops being as light, so the whole photograph occupies a smaller slice of the available range. A scan of it is a perfectly accurate record of a picture with the contrast squeezed out.
That is a recoverable condition, and it is recoverable arithmetically rather than by judgement, because the missing range is measurable. Read the histogram of each channel, find where its data really begins and ends, and map that back onto nought to two hundred and fifty-five. Nothing is invented; the levels that survived are simply spread back out over the space they used to occupy.
| Channel | Range still held | After recovery |
|---|---|---|
| Red | 18 to 249, so 231 levels | Barely moved — this channel was fine |
| Green | 34 to 236, so 202 levels | Stretched, mid-tones lift slightly |
| Blue | 61 to 198, so 137 levels | Stretched hardest — this is where the orange goes |
Why per channel, and not one curve for all three
One curve applied to all three channels keeps their relationship to each other exactly as it is, which means it keeps the cast. It makes a faded orange print into a contrastier orange print. Measuring separately is what lets the blue channel — which lost the most — be stretched the most, and the cast comes off as a side effect of treating each layer according to what it actually lost.
What it will not do
A channel with almost nothing left in it is not recoverable. If a channel holds fewer than eight levels of range, the correction leaves it alone rather than multiplying a handful of values by thirty and calling the noise a picture. You will see this on badly light-struck prints, where one corner has gone and the rest has not, and it is the correct outcome: the corner has no data, and a tool that pretended otherwise would be selling you a guess.
A scan worth correcting
Fade recovery is only as good as the range in the scan it is handed. A scanner that auto-corrects on the way in has already made these decisions, usually badly, and has clipped whatever it decided was black. Switch every automatic option off and scan flat. A flat scan looks worse on the scanner’s preview and gives this page more to work with, which is the trade worth making. The scanning piece goes through the rest of it.
Questions about faded prints
- How old does a print have to be before it needs this?
- Age is the wrong variable. What matters is what the print was made of and where it was kept. A resin-coated colour print from 1985 left in a south-facing room can be further gone than a properly fixed black-and-white print from 1930 kept in a drawer. The measurement tells you: open the scan, turn fade recovery to full, and read the report. If it says red held 240 levels of 255 and blue held 90, the print has faded badly whatever its date.
- Why is my old print orange rather than simply pale?
- Because the three dyes do not fail at the same rate. In most colour processes of the 1960s to 1980s the cyan dye is the least stable, and when cyan goes the red and yellow left behind read as orange. That is why a global warmth correction is the wrong tool here and per-channel measurement is the right one — the picture is not uniformly too warm, it is missing one specific layer's worth of range.
- The recovered picture looks grainy and blotchy in the sky. Why?
- Stretching a channel stretches whatever is in it, and a channel that has faded to ninety levels of range has ninety levels of signal and the same amount of scanner noise it always had. Pull the fade slider back to sixty or seventy per cent. A partial correction that keeps the grain quiet is usually a better photograph than a full one that does not, and the slider previews live so you can find that point by eye.
- Should I fix the fading before or after the scratches?
- The engine does it in one order and the order is fixed: damage is marked, the marked pixels are filled, and the tone curves are applied last. That is deliberate. Endpoint measurement excludes every marked pixel, so a bright scratch cannot set the white point, and filling before toning means the filled pixels get the same curve as everything around them rather than standing out under it.
- Can I keep the warmth? It looks like the period it came from.
- Yes — that is what the slider is for rather than a checkbox. Fade recovery at thirty or forty per cent brings the contrast back while leaving most of the cast in place. The density control underneath it is the other half of that judgement: positive values make the picture denser, which is what a darkroom print of the era would have looked like next to a flat scan of a faded one.