What the grain keeps
There is a slider in every retouching program, it bears a name that leaves no room for doubt, noise reduction, and most people push it all the way without thinking about it for a second. Which is logical. Nobody wants to make an image dirty, nobody wants to hear that it lacks sharpness, and the word noise says clearly enough that this is a problem to be made to disappear.
Except that this slider does not remove a flaw, it removes information. And at the end of the run, once you have pushed it to the maximum, what remains is not a clean image, it is a plastic image, those smoothed-over faces you recognise immediately without being able to say why they ring false.
Grain is the skin of the image. You can smooth it, but you cannot replace it.
THE MATERIAL
Let us start with what it really is, because that is where the confusion comes from. On photographic film, the image is carried by silver halide crystals suspended in a layer of gelatin. These crystals come in irregular sizes, they are scattered at random, and above all they are scattered through the thickness of the emulsion, not across a surface. The film's sensitivity depends directly on their size; the bigger the crystals, the faster they catch the light, and the more they show. That is why a very fast film is visibly grainier; it is not a manufacturing defect, it is the accepted price of speed.
What this means is that grain is not laid on top of the image. It is the image. There is nothing underneath, no smooth version of which the grain would be the degradation; the film image exists only as a cluster of silver particles of different sizes.
It is not noise added to a signal. It is the signal itself, seen up close.
THE MISTAKE
Digital noise has nothing to do with it, and it is this misunderstanding of vocabulary that skews everything. It comes from two sources. The first is physical and impossible to remove; photons do not arrive on the sensor in a regular flow, they arrive statistically, so that a single uniform grey does not receive exactly the same amount of light from one photosite to the next. The second is electronic, it is what the sensor and its amplification add along the way.
The essential difference with film is that all of this happens on a perfectly regular grid. The photosites are aligned, identical in size, in a plane. The noise therefore settles into a lattice, and the eye, which spots regularities better than anything, reads it as an anomaly laid over the image, whereas it reads film grain as a material.
And while we are at it, a word about what that sensor actually produces. Since the patent filed by Bryce Bayer at Kodak in the mid-seventies, almost all sensors are covered with a mosaic of coloured filters, half green, a quarter red, a quarter blue. Each photosite therefore measures only one colour, and the other two are deduced by calculation from its neighbours. A digital photo is in very large part interpolated. It is less measured than we think.
THE PROOF
Where it gets interesting is when we leave the technique behind and ask why grain does this to us.
In Camera Lucida, published in 1980, Roland Barthes looks for what fundamentally distinguishes photography from every other mode of representation, and he ends up condensing it into a formula that went around the world, the that-has-been. Photography does not say that something looked like this, it says that something was there, in front of the lens, and that the light made the journey.
Grain is the physical trace of that journey. Each silver crystal reacted or did not react, there is no third possibility, and what we are looking at is the record of that encounter between photons and chemistry. A smoothed image erases that record. It remains accurate, it remains a likeness, but it has lost the proof that a contact took place.
And that, the eye knows before we do.
THE RETURN
Hence the paradox that amuses me most in this story. The industry spent forty years removing grain, and it now spends considerable time putting it back.
In cinema, adding grain in post-production has become a step in its own right, to the point that you can buy banks of grain scanned from real film stock. Video games offer a film grain setting in their options. The most used filters on photo apps are the ones that simulate vanished emulsions.
But the example that seems most telling to me comes from sound, because it is technical and leaves no room for the argument of taste. When you reduce the resolution of an audio file, the rounding of values creates a distortion correlated with the signal, particularly unpleasant to the ear. The solution used for decades consists in deliberately adding a very low level of noise before the conversion, what is called dither. You measurably degrade the signal-to-noise ratio, and the result is better. The same principle applies to image gradients, where visible banding is broken up by injecting noise.
In other words, in two different trades, the clean solution produces a dirty result, and the dirty solution produces a clean one.
THE PRICE
There remains what happens in real life, on a project, with someone across the table. A client never asks for grain. They ask for sharpness, for quality, for high-end, and those three words, in their mind, mean the same thing, an image without asperity.
This is not ignorance, it is an inheritance. Throughout the twentieth century, grain signalled a lack of means, cheap film, insufficient light, the amateur. It took photographers willing to publish images their era judged technically failed for that sign to change meaning and become, today, the sign of authenticity.
What the grain keeps is therefore not only material. It is the memory of how the image was made, and the one detail that still prevents us from believing it was fabricated out of nothing.
A silver crystal that reacted, a photon that arrived, an emulsion of a given thickness, a noise added on purpose so that it sounds right. What all of these have in common is that none of them is decorative, and that they disappear at the first slider pushed a little too far.
So when I leave grain in, it is not nostalgia, nor a filter, nor a stylistic effect fallen from a fashion. It is what remains of the encounter between a light and a surface, and it is exactly what makes the difference between an image you look at and an image you believe.
Grain is not what dirties the image. It is what proves it took place.
To dig further. Roland Barthes, Camera Lucida, for the that-has-been. Bryce Bayer's patent filed at Kodak in 1976, to understand what a sensor measures and what it guesses. And any audio mastering manual for the question of dither, which explains better than I can why noise gets added on purpose.
