Grain comes out, and so does the skin texture

Grain is not blur and it is not low resolution

These three get treated as one problem called bad quality, and they need opposite treatments, so the first job is naming which one you have. Zoom to full size on a flat area, a wall, a sky, a cheek, and look at what is wrong with it.

Grain, or noise, is speckle. Random light and dark dots across an area that should be even, often with coloured flecks mixed in. It is strongest in the shadows and in anything shot in a dim room. The fix is denoising, which is this page.

Blur is smear. Edges that should be sharp are spread out in a direction, or everything is soft because focus landed somewhere else. No amount of denoising helps, and denoising makes it worse by removing what little edge detail survived. That is a different job, covered by the guide on unblurring a picture.

Low resolution is blockiness. Not speckle and not smear but visible squares, soft steps along a diagonal line, or the mushy edges that come from an image being enlarged past what it holds. The resolution guide handles that, and compression damage from re-saving has its own guide too, because JPEG artefacts look like grain and are not grain.

Getting this wrong is the common waste. People apply noise reduction to a soft photo, see it get softer, and conclude the photo is unfixable, when the actual defect was never noise.

Why the grain is there in the first place

Noise is not dirt that landed on the picture. It is the sensor reporting its own uncertainty, and knowing which kind you have tells you how much is recoverable.

Not enough light is the usual cause. In a dim room the sensor collects too little signal to be confident about each pixel, and the camera raises its sensitivity to compensate, which amplifies the uncertainty along with the picture. This is why indoor and night photographs are grainy and daylight ones are not, on exactly the same phone.

A small sensor makes it worse at the same light level, which is why a phone photograph of a dim room is grainier than a camera photograph of the same room. There is nothing to fix in that, it is the physics of a small lens and a small sensor.

Brightening afterwards is the cause people miss. A photograph that was underexposed and then lifted in editing gains grain in the process, because raising the shadows raises the uncertainty that was hiding in them. The grain was always there, the brightening just made it visible, which is also why fixing a dark photo and fixing a grainy one tend to turn into the same job.

Scanned prints and film are a separate case. Film grain is part of the photograph rather than an error in it, and the dust and speckle on an old scan is a surface problem. Treating either as sensor noise flattens the image and loses what made it look like a photograph of its period.

What denoising actually takes away

The software is making a judgement about every small region: is this variation the subject or is it noise? Where it guesses wrong it smooths something real, and the things it gets wrong are consistent and worth knowing by name.

• Skin stops looking like skin. Pores and fine lines are exactly the scale of noise, so they go first, and the result is the waxy or plastic look. It is the single most common sign that denoising went too far, and faces are where any viewer notices it immediately.
• Hair turns into shapes. Individual strands merge into soft masses, and the edge between hair and background goes smooth in a way that reads as a cut out rather than a photograph.
• Fabric loses its weave. Knitwear, linen and anything woven become flat colour, which takes the sense of material out of the picture even when nothing looks obviously wrong.
• Fine text and thin lines smear. Anything with lettering in it, a sign, a label, a document, suffers earlier than the rest of the frame.
• Colour patches appear. Aggressive colour noise reduction can leave broad flat areas of slightly wrong colour where it has averaged speckle away, most visible in skies and shadows.

How to tell you have gone too far

The mistake is judging a denoise on the whole image at screen size, where everything looks better because all small detail reads as cleaner. Judge it the opposite way.

• Zoom to full size on a face before anything else, and look only at the skin and the eyelashes. If the skin has lost its texture or the lashes have merged into a dark band, the setting is too high, whatever the rest of the frame looks like.
• Compare against the original rather than against the previous attempt. Working upwards in steps makes each step look reasonable and hides how far the whole sequence has travelled.
• Check one area that should be flat and one that should be detailed. A good setting leaves the flat area clean and the detailed one still busy. A bad one makes both equally smooth.
• Accept visible grain in exchange for real texture. A photograph with some grain still in it reads as a photograph. A photograph with no grain and no texture reads as a rendering, and viewers register that as fake without being able to say why.
• Keep the original file untouched, always. Denoising is the one edit that genuinely cannot be undone from the saved result, because the detail it removed is not in there any more.

Stopping it at the camera, which is cheaper than fixing it

Grain is one of the few photograph problems that is substantially easier to prevent than to repair, and the preventions are not technical.

• Add light rather than brightness. Opening a curtain, turning on a lamp or moving somebody nearer a window does more for the final file than any setting on the phone.
• Hold still and let the camera choose a longer exposure instead of a higher sensitivity. Bracing against a wall or a table is enough, and it is the trade most phones will take if the shot is steady.
• Expose for the subject rather than planning to lift the shadows later. A photograph taken slightly bright and pulled down keeps far more than one taken dark and pushed up.
• Take several frames of anything important in poor light. Noise is random, so the best of a handful is noticeably better than any single attempt, at no cost.
• Avoid digital zoom in a dim room, which crops into the grainiest part of what the sensor managed to collect and then enlarges it.

Letting Clara make the trade for you

The reason this is hard by hand is that the right amount of denoising is different in every part of the frame, and a single slider applies one decision everywhere. Clara handles a grainy photograph as a whole: you hand it the file and it comes back cleaned up, with the detail that matters kept rather than averaged away, and no panel to balance.

Judge it the same way you would judge your own attempt. Open the result at full size, look at a face, and check the skin and the lashes against the original before you keep it. Clara shows you the before and after side by side for exactly that reason, and a result that has lost the texture of a face is not a result worth saving, whoever produced it.

What it cannot do is invent what the sensor never recorded. A photograph taken in near darkness has little real information in the shadows, and anything smooth that appears there is a guess rather than a recovery. Grain over a real face beats a clean surface where the face used to be, and that is the honest limit on all of this. The guide on improving photo quality covers the cases where the defect turns out to be something other than noise.

More guides

• how to improve photo quality
• how to unblur a picture
• how to increase image resolution
• why iphone photos come out blurry
• jpeg artifacts
• how to brighten a dark photo
• how to fix an overexposed photo

Questions