How to Fix Grainy Night Photos (and Avoid the Grain Next Time)

Night photos are where phone cameras stop looking like magic. The bar looked warm and cinematic, the city looked like a postcard, the birthday candles looked perfect — and the photo came back speckled, mushy, with skin that looks like sandpaper and shadows crawling with coloured specks.

That speckling is noise, and it is worth understanding before reaching for a tool, because half the fix happens before you press the shutter and the other half has real limits. This guide covers both, honestly.

Why night photos come out grainy

Light is the raw material, and at night there is very little of it

A camera sensor works by counting photons. In daylight, each pixel on the sensor collects a huge number of them, so the reading is confident and stable. At night, each pixel collects very few — and the arrival of light is inherently random. When the count is low, that randomness becomes visible as speckle. Photographers call it shot noise, and it is not a defect in your camera. It is physics.

Phone sensors feel this harder than big cameras do, because each of their pixels is tiny and collects even less light. A photo of the same dim room taken on a phone and on a large-sensor camera will differ mostly in how much of that randomness you can see.

ISO amplifies the noise along with the picture

When there is not enough light, the camera turns up the gain — that is what raising ISO means. Gain does not create signal. It multiplies whatever the sensor recorded, and the random speckle gets multiplied by exactly the same amount. Push it far enough and you also get colour noise: red, green and blue blotches in shadow areas where the sensor was mostly guessing.

Phones hide some of this with computational photography, stacking several frames to average the randomness out. That is why night mode helps, and why it fails the moment anything in the frame moves.

Compression adds a second kind of grain

There is a second, sneakier source. When a noisy photo gets saved as a JPEG — and again when it is sent through a chat app, and again when a platform re-encodes it — the compressor has to describe all that random speckle, which it is very bad at. What comes back has blocky patches, smeared colour and a mottled texture that was not in the original noise. Many photos people describe as "grainy" are really a noisy shot that has been re-compressed several times.

Prevent the grain when you shoot

The cheapest denoise is the one you never need. In rough order of impact:

1. Hold still, or find something to lean on. A steadier camera lets the phone choose a longer exposure, which collects more light, which means less gain and less noise. A doorframe, a table or a railing works. So does simply bracing your elbows against your body.
2. Use night mode and let it finish. Night mode captures a burst and merges it. Moving the phone — or letting your subject move — during that window is what produces the smeary, half-sharp results people blame on the mode itself.
3. Do not use digital zoom. Pinching to zoom crops into the sensor, so you keep less of the light that was already scarce. Walk closer if you can. If you cannot, shoot wide and crop later, which at least keeps every pixel the sensor recorded.
4. Put your subject near the light that exists. A face lit by a shop window, a streetlight or a candle will always beat a face lit by nothing, and moving one step is faster than any editing.
5. Expose for the subject, not the scene. Tap the person's face on the screen so the camera meters for them. Photos brightened afterwards in an editor amplify the noise in exactly the way raising ISO does.
6. Avoid the extra round trips. Send the original file rather than a screenshot, and pull photos off the phone directly rather than forwarding them through apps that re-compress on the way.

What can be done about a grainy photo you already have

This is where honesty matters more than marketing.

What AI enhancement actually helps with

ImageEnhance runs an AI super-resolution pipeline. Its job is reconstruction: it reads the picture, works out what structure is likely present, and rebuilds detail at up to four times the original resolution. That pipeline powers both the enhance and the upscale tools — they are the same engine.

For night shots, that produces a real but specific set of improvements:

• Light grain gets tidied as a side effect. The model rebuilds edges and textures from what it recognises, so fine random speckle tends not to survive reconstruction. It is not targeting the noise; it simply does not redraw it.
• Detail buried under noise can come back. Hair, fabric, lettering on a sign, the edge of a jaw — features that the speckle was drowning often reappear with usable definition.
• Softness improves. Many night photos are not only noisy but slightly soft, because the longer exposure blurred things a little. Reconstruction helps there too.
• Enlarging stops making it worse. Stretching a noisy photo the ordinary way magnifies every speck. Rebuilding it does not.

What it will not do

• Heavy noise is not reliably removed. When speckle is dense, the model can read it as genuine texture and rebuild it faithfully. A photo shot in near-darkness may come back sharper and still grainy.
• Colour blotches in deep shadow are the hardest case. There is little real signal underneath them to reconstruct from.
• Motion blur is a different problem. If your subject moved during the exposure, the information about where they actually were was never captured.

The realistic expectation: a mildly grainy night shot usually comes back clearly better. A photo that was already at the edge of usable comes back a bit better. Since running one is free, the fastest way to find out which yours is, is to try it.

A dedicated denoise tool is coming

Smart denoise, old photo restoration and AI colourization are marked coming soon on the site, and that label is literal — they are not live yet. What exists today is the enhance and upscale pipeline described above. When a purpose-built denoise model ships, it will target grain directly instead of cleaning it up as a by-product, which is a meaningfully different thing.

Step by step: run a night shot through the enhancer

1. Pick the best copy of the photo. The original file off the camera roll, not a screenshot and not the version a friend forwarded back to you. Fewer compression rounds means less of that second kind of grain.
2. Crop before you enhance. If you only need the two people and not the whole dark street, crop first — the reconstruction then spends all its effort inside the frame you are keeping.
3. Upload it. Go to ImageEnhance, drag the file in or paste it from the clipboard. JPG, PNG and WebP are supported, up to 4MB.
4. Choose the output size. 2K output is free — you do not need an account to run a photo. 4K output comes with a subscription starting at $8 a month; see the pricing page.
5. Compare at full size. Drag the before/after slider and look at faces and flat areas like walls and sky, which is where noise shows most. Judge it at full size, not on a thumbnail.
6. Download the result. A 2K result typically takes a few seconds; 4K about ten.

Without an account you get 8 free credits every 24 hours, enough for two 2K enhancements — so trying a couple of candidate photos costs nothing. Signing up adds 30 credits on the spot, then 10 free credits every day after that.

FAQ

Can AI completely remove noise from a night photo? Not today, and not honestly. The current pipeline is super-resolution, and it reduces light grain as a consequence of rebuilding detail rather than by targeting noise. Dense noise can survive, or be rebuilt as texture. A dedicated denoise tool is listed as coming soon.

My photo is grainy and blurry. Which is fixable? Softness and light grain generally improve. Motion blur — where the subject moved during the exposure — does not, because the detail was never recorded in the first place.

Should I brighten a dark photo before enhancing it? Usually not. Brightening amplifies noise in the shadows exactly the way raising ISO does, and hands the model more speckle to interpret. Try the photo as it is first.

Does this work on phone photos? Yes — phone shots are the most common input, and the whole thing runs in a browser on desktop or mobile. JPG, PNG and WebP up to 4MB, with the result rebuilt at up to four times the original resolution.

Got a night shot you were about to delete? Try it first — it costs nothing to find out.