How to Fix Grainy Photos: RAW Denoise Guide (2026)

Published August 31, 2026 · Daniel Tello · RevelRaw

You open a RAW from a dim reception or a night street walk and the frame that looked fine on the camera screen has turned to colored speckle. The composition holds. The shadows look dirty, skin has blotches, and every pass of sharpening makes it worse. This guide covers where the grain comes from, the order of operations that removes it without turning faces to wax, when AI denoise earns its keep over manual sliders, and how to tell a fixable file from a reshoot.

Why photos turn grainy

Grain is mostly decided before the file ever reaches an editor. Low light gives the sensor few photons to measure. High ISO amplifies that weak signal and every flaw riding on it. Small pixels sit close to their own noise floor, and an underexposed frame turns noisy the moment you lift it. Marc Levoy's Stanford lecture notes on noise put the capture order plainly: open the aperture as far as depth of field allows, slow the shutter as far as motion allows, and raise ISO only after both.

• Shoot the lowest ISO the shutter can afford. A sharp, noisy frame survives editing; a clean, blurred one does not. Never buy a lower ISO with motion blur.
• Expose toward the bright side of the histogram. Protect true highlights, but stop leaving the subject two stops down for safety. Starved shadows are where grain lives.
• Add real light when the scene allows it. A faster lens, a tripod for a static subject, a bounced flash or a step toward the window all feed the sensor photons that no slider can add later.

A well-exposed high-ISO file still carries texture, and that is fine. You are aiming for controlled texture that leaves detail alone, not a sterile image.

Luminance noise and color noise are different problems

At 100 percent, grain separates into two defects. Luminance noise reads as monochrome speckle, a rough texture laid over the image. Color noise reads as red, green and blue blotches, worst in shadows, dark clothing and dark skin tones. They deserve different treatment. Color blotches rarely carry real information, so strong chroma reduction costs little. Luminance speckle overlaps genuine texture in hair, fabric and foliage, so it deserves a lighter hand.

The physics explains why shadows suffer first. Photon arrival is statistical, and its variation shrinks only with the square root of the light collected, so a dark pixel holds proportionally more uncertainty than a bright one. Imatest's noise documentation measures the payoff in the high-ISO case, where sensor noise dominates: signal-to-noise ratio improves by about 6 dB for each doubling of exposure. Light captured in camera builds a stronger file than the same weak signal brightened afterward.

Why editing makes grain worse before it makes it better

A RAW editor receives no extra photons when you raise exposure or recover shadows. It stretches what the sensor recorded, unwanted variation included, and the Stanford notes warn that brightening in post enhances noise more than raising ISO at capture would have. The grain you see after a three-stop shadow lift was in the file all along.

That settles the editing order: correct exposure and white balance, denoise, then sharpen. Sharpening a noisy file treats every speckle as an edge and makes the grain more confident. This is also where shooting RAW pays off, because a JPEG has already had its shadows quantized and its noise baked in; the RAW vs JPEG comparison shows how much further a lifted RAW shadow goes.

A denoise workflow that holds up

1. Inspect at 100 percent. Check skin, hair, fabric, foliage and the deepest shadow that matters. A fit-to-screen preview hides color blotches and forgives oversmoothing; full magnification forgives nothing.
2. Set exposure and white balance first. Denoise should respond to the version of the photo you intend to keep. A wrong color cast makes chroma noise impossible to judge.
3. Reduce color noise first. Push it until the blotches settle. You will be surprised how far it goes before anything real disappears.
4. Reduce luminance noise gently. Move the control until the largest distracting speckles calm down, then stop before fine texture follows them. If your editor offers local masks, spend the strong cleanup on walls, skies and out-of-focus backgrounds; with global controls, let the subject decide the ceiling and leave the background slightly noisier.
5. Compare before and after at the same zoom. A split view or a toggle keeps you honest about what the cleanup cost. Waxy skin and melted foliage announce themselves side by side.

Standard denoise or AI denoise

Manual luminance and chroma sliders are predictable. They apply the same medicine to every pixel, which suits mild noise, batches that need consistent treatment, and photographers who want to decide exactly how much texture remains. Their weakness is that a global filter cannot tell an eyelash from a speckle, so pushed hard enough to clean a dark background, it smears the detail you kept the frame for.

AI denoise works from the other direction: a trained model estimates which structures are edges and texture and which are noise, and cleans around the difference. The research backs the approach honestly rather than breathlessly. The RENOIR benchmark tested denoisers on real low-light photographs, over 100 scenes and more than 400 images, and found meaningful quality differences between methods rather than a free lunch. The Darmstadt Noise Dataset paper makes the stage argument: the same noisy captures measured 39.39 dB PSNR in linear RAW against 29.98 dB after conversion to sRGB, which is why denoising belongs in the RAW pipeline, not on the exported JPEG.

RevelRaw takes the AI route without the cloud: its AI Denoise runs a neural model on your Mac's own chip, offline, clearing luminance and color noise at full resolution while keeping the fine texture you shot for. You set the strength, and your photos never leave the machine. For difficult shadows and strong high-ISO noise it is the better starting point; for a mildly noisy batch that needs identical treatment, manual sliders remain the predictable tool.

Sharpen after, and sharpen selectively

Noise reduction softens fine edges a little, and sharpening buys them back, but only in that order. Start conservative: on a 100-point amount slider, 25 to 40 is a sane opening bid, with a small radius for densely detailed files. A denoised photo usually needs less capture sharpening than a clean one, so reduce the default instead of trusting it.

If your editor has a sharpening mask, raise it until the sharpening concentrates on real edges instead of the flat areas where noise lives, and use the Alt or Option preview if one exists: white receives sharpening, black is protected. Then check the usual witnesses at 100 percent. Texture crawling across skin or sky means the mask is too permissive. Bright halos around faces and signs mean too much amount or radius. Save output sharpening for the destination, and pick it per medium; the Instagram export guide covers the small-screen case.

Three scenes, three balances

Concert photography

Deep shadow, colored stage light and moving subjects. Calm the red, green and blue speckle with chroma reduction first, keep luminance reduction moderate so faces and instrument edges stay believable, and inspect the saturated highlights before export, because colored lighting hides blotches a neutral scene would show. Black backgrounds tolerate strong cleanup; eyes and microphone grilles do not.

Indoor weddings

Mixed tungsten and window light makes noise look worse than it is, because color casts and chroma blotches overlap. Correct white balance before judging anything, then clean gently and keep skin dimensional. A smooth wall can go sterile; a veil, a bouquet and a dress hem need their texture back at the end.

Night street scenes

Large dark areas, bright signs, and motion you cannot reshoot. Keep the shadow lift restrained before denoising, protect the clean edges around neon and lettering when sharpening, and leave reflections alone: wet asphalt shimmer looks like noise at a glance and reads as life in the print. Capture settings decide most of these files, and the free EXIF viewer shows the recorded ISO, shutter and aperture, which tells you whether the softness you see is noise, motion or focus.

When to edit and when to reshoot

Denoise improves a difficult file. It cannot restore what was never captured, and the triage takes a minute at 100 percent:

• Is the defining detail present? Visible eyes, lettering or subject edges mean the file is worth the work.
• Is the blur motion? No noise tool reverses a subject that moved during the exposure.
• Are the shadows noisy or empty? A textured shadow cleans up. A void lifts into gray mush with nothing to recover.
• Are important highlights clipped? With no RAW headroom left, lowering them restores brightness, never texture.

Moderate high-ISO noise over a clean exposure and strong subject edges: edit. Deep underexposure plus motion blur plus clipped highlights: reshoot if another take exists. Before delivery, review the denoise and sharpening together at export size, check saturated areas for leftover blotches, gradients for banding, and skin for wax. Grain you add back on purpose is a different subject, and the film-look guides cover why chosen grain over a clean file works where noise never does.

FAQ

Why do my photos look more grainy after editing?

Raising exposure or lifting shadows stretches the signal the sensor recorded, noise included. The grain was in the file all along; the edit made it visible. Correct exposure first, then denoise the version you intend to keep, so the cleanup responds to the real problem instead of a moving target.

Can you fix a grainy photo without losing detail?

Mostly, when the detail is there to protect. Reduce color noise first, since the blotches rarely carry real information, then reduce luminance noise until the largest speckles settle and stop before fine texture goes waxy. No tool rebuilds detail the sensor never captured: empty shadows and clipped highlights stay empty.

Should I sharpen before or after noise reduction?

After. Sharpening treats every speckle as an edge, so sharpening first makes the grain more prominent. Denoise, then sharpen conservatively, around 25 to 40 on a 100-point amount slider, and check skin and skies at 100 percent for texture that should not be there.

What is the difference between luminance noise and color noise?

Luminance noise is monochrome speckle, a rough texture over the image. Color noise is red, green and blue blotches, worst in shadows and dark saturated areas. Color noise takes strong reduction well; luminance noise overlaps genuine texture in hair, fabric and foliage, so it deserves a lighter hand.

Does AI denoise upload my photos to a server?

Some cloud tools work that way. RevelRaw's AI Denoise runs its neural model on your Mac's own chip, offline and at full resolution, so your RAW files never leave the machine. You set the strength and compare the result against the original in the split view.

Related reading: RAW vs JPEG on a Sony Alpha · The film look from Canon RAW · The Kodak Portra preset for Mac RAW photos · Lightroom alternatives without a subscription