Reduce Image Noise Online
A median filter that removes speckle and high-ISO colour noise while leaving edges sharp — unlike a blur, which smears noise around and softens everything. Colour-only mode keeps detail intact, and large photos show real progress instead of freezing the tab. Nothing is uploaded.
Drop images here
or click to browse · or paste with Ctrl+V
JPG · PNG · WebP · GIF · BMP · AVIF
What to clean
Blend between the original and the filtered result.
Compares 9 pixels. Higher removes more noise and more texture, and costs more time.
Presets
Your images are never uploaded. The filter runs inside this browser tab.
People also use
Everything this tool does
Median, not a blur
Takes the middle value of each neighbourhood and discards outliers, rather than averaging them in. On impulse noise it removes what a blur only spreads around.
Edges survive intact
In testing, a hard edge came through the median filter with its full contrast, while a blur of the same radius softened it by more than 20 percent.
Colour-noise-only mode
Smooths chroma and puts the original brightness back, so the red and green speckle of high ISO disappears while texture and detail stay.
Speckle mode
Only replaces pixels that differ sharply from their neighbours, leaving the rest of the image byte for byte identical. Ideal for dust on a scan.
Real progress, no freeze
Full-resolution work is done in row bands with the browser given control between each, so a 24 megapixel photo reports progress instead of hanging the tab.
Drag-to-compare preview
A split handle wipes between original and cleaned, which is essential here because the difference between clean and plastic is easy to miss.
Strength as a blend
Rather than an on-off filter, strength mixes between original and result, so you can leave a little natural noise instead of erasing all texture.
Batch with one ZIP
Apply the same settings across a set shot at the same ISO and download everything as a single archive.
Transparency preserved
The alpha channel is untouched for PNG and WebP output, so only visible pixels are filtered.
Nothing is uploaded
Noisy photos are usually private low-light ones. Nothing leaves your device, and there is no server copy to delete.
Median versus blur, measured
These are results from our own test on a flat grey patch with salt-and-pepper noise and a hard vertical edge, both filters at the same radius. It is the clearest way to see why the choice of filter matters more than the settings.
| Measurement | Original | Median (this tool) | Blur, same radius |
|---|---|---|---|
| Average deviation from flat grey | 11 | 0 | 10 |
| Hard edge contrast retained | 130 | 130 | 102 |
| Isolated bright pixels remaining | many | none | spread out |
| Fine texture | intact | reduced at high radius | reduced everywhere |
Honest caveat: a median is a classical filter, not an AI denoiser. On very heavy noise a trained model will do better. This runs instantly, privately and with no model to download.
3
cleaning modes
49
pixels compared max
0
files uploaded
Live
progress reporting
ZIP
batch download
Free
no account needed
FlipMyFormat vs other denoise tools
| Capability | FlipMyFormat | Typical free tool |
|---|---|---|
| Files stay in your browser | ||
| Median filter, not a disguised blur | ||
| Colour-noise-only mode | ||
| Speckle-only mode that skips clean pixels | ||
| Real progress instead of a frozen tab | ||
| Batch ZIP download without paying | ||
| No watermark on the output |
How to reduce noise without losing detail
The trade-off in every denoiser is the same: noise and fine texture look almost identical to an algorithm, because both are small, rapid variations between neighbouring pixels. Push the strength up and you do get a clean image, but skin loses its pores, fabric loses its weave and foliage turns into green mush — that waxy, plastic look is not a bug, it is what happens when you ask a filter to remove all small variation. The useful escape from that trade-off is to notice that the two are not equally distributed across the channels. Digital sensor noise at high ISO is overwhelmingly chroma noise: those red and green speckles sitting in the shadows carry no real information at all, while the brightness variation is largely genuine texture plus a grain that most people find acceptable, even attractive. Colour-only mode exploits exactly that — it smooths the colour and then restores the original brightness pixel for pixel, so the speckles vanish and the detail is mathematically untouched. That is why it is the default here and why it should be your first attempt on almost any high-ISO photo. Two other things follow from the same reasoning. Denoise before sharpening, never after, because sharpening amplifies noise and makes it permanent. And keep the radius low: it is tempting to raise it when noise persists, but the cost climbs steeply — radius 3 compares 49 pixels per output pixel instead of 9 — and so does the texture loss. Strength is the safer dial, and leaving a little noise almost always looks better than erasing everything.
- 1
Add your images
Drag them in, click to browse, or paste with Ctrl+V. Denoise before any sharpening or resizing.
- 2
Start with colour only
Try colour-noise-only mode at radius 1 first. Raise strength before radius, and watch the compare handle for texture loss.
- 3
Apply and save
Press Clean. Large photos show real progress. Then download individually or take the batch as one ZIP.
How the filter works
Why the middle value wins
For each pixel the tool collects its neighbours, sorts them, and takes the middle one. A single wildly wrong value — a hot pixel, a dust speck, a chroma speckle — sorts to one end of that list and is never selected, so it disappears completely. An average cannot do this: the outlier is included in the sum and drags the result, which is why blurring redistributes noise rather than removing it. The same property is what preserves edges: on one side of an edge the majority of neighbours are dark, on the other the majority are light, so the median stays firmly on the correct side.
Why it needs a progress bar
Sorting a window for every pixel is expensive, and the cost grows with the square of the radius. On a 24 megapixel photo at radius 3 that is roughly 1.2 billion comparisons, which a single uninterrupted JavaScript loop would spend around half a minute on with the tab completely frozen. Instead the image is processed in bands of rows, and after each band the tool yields control back to the browser before continuing. The page stays responsive, the progress bar reflects genuine completed work, and nothing is faked to look faster than it is.
What people use it for
Low-light phone photos
Indoor and night shots are full of chroma speckle. Colour-only mode cleans them without turning faces into plastic.
High-ISO event photography
Concerts and weddings shot at ISO 6400 clean up well, and the file often shrinks noticeably too.
Scanned prints and negatives
Removes paper grain and small dust specks in one pass. Speckle mode leaves everything else untouched.
Before sharpening
The correct first step in any workflow, because sharpening noise makes it permanent.
Astrophotography previews
Sensor hot pixels are exactly the isolated outliers a median discards completely.
Cleaning up screenshots
Screenshots of video or compressed streams carry blocky mottling that a light median softens.
Reducing web file size
Noise is incompressible, so removing it can cut 20 to 40 percent off a JPEG at the same quality setting.
Product shots in poor light
Cleaner backgrounds look more professional on a listing page, and colour-only mode keeps material texture.
Six tips worth knowing
Try colour-only mode first on any high-ISO photo. It removes the speckle that actually bothers you and leaves detail mathematically untouched.
Raise strength before radius. Radius costs time and texture; strength only costs a blend, and it is much easier to judge.
Denoise before sharpening, always. Sharpening noise bakes it in and no later filter fully removes it.
Leaving a little noise usually looks better than erasing all of it. Fully clean images read as artificial to most viewers.
For dust on a scan, use speckle mode at full strength. It only touches isolated outliers, so the rest of the scan is untouched.
Check the file size afterwards. Removing noise often cuts 20 to 40 percent off a JPEG, which is a free performance win for web use.
Troubleshooting
Skin looks waxy and plastic
Strength or radius is too high in a mode that touches brightness. Switch to colour-only, which cannot remove luminance texture at all.
It is taking a very long time
Radius 2 and 3 are far more expensive than radius 1, and the cost scales with megapixels. Lower the radius; the progress bar is real work, not a stall.
The noise is still there
If it is brightness grain rather than colour speckle, colour-only mode will not touch it by design. Switch to colour and brightness mode.
Fine detail disappeared
The filter cannot distinguish fine texture from noise at higher radii. Drop to radius 1 and raise strength instead.
Frequently asked questions
How do I reduce noise in a photo online?
Drop your image onto the tool, choose a mode and set radius and strength. The preview updates live and the compare handle wipes between original and cleaned. Press Apply and a progress bar runs while the full-resolution version is processed. Nothing is uploaded — it all happens in your browser.
Why use a median filter instead of a blur?
Because a blur averages a noisy pixel together with its neighbours, which spreads the noise across a wider area rather than removing it, and softens every edge in the process. A median takes the middle value of the window and discards outliers entirely. In our own testing on a flat grey patch with salt-and-pepper noise, the median reduced the average deviation to zero while a blur of the same radius only got it from 11 to 10 — and the median left a hard edge completely intact where the blur softened it by more than 20 percent.
What do the three modes do?
Both cleans brightness and colour together, which is the strongest option and the most likely to soften fine texture. Colour only smooths the colour information while putting the original brightness back afterwards, so noise disappears but detail survives. Speckle only replaces a pixel when it differs sharply from its neighbours, so most of the image is left byte for byte identical and only the outliers change.
Which mode should I use for a high-ISO photo?
Colour only, almost always. Digital sensor noise at high ISO is mostly chroma noise — those red, green and blue speckles in the shadows — while the brightness variation reads as acceptable grain. Removing the colour speckle and keeping the luminance grain gives a result that looks clean without looking like plastic.
What does radius change?
How large a neighbourhood each pixel is compared against. Radius 1 looks at 9 pixels, radius 2 at 25, radius 3 at 49. Larger radii remove more noise but also erase genuinely fine texture, and they cost significantly more time. Start at 1 and only raise it if the noise is clearly still there.
Why does it show a progress bar?
Because a median is genuinely expensive: every pixel requires sorting its whole neighbourhood. At radius 3 on a 24 megapixel photo that is nearly 1.2 billion comparisons. Rather than freeze the tab for half a minute, the tool processes the image in bands and hands control back to the browser between each one, so the page stays responsive and you can see how far along it is.
How long will a large photo take?
Roughly one second for a 6 megapixel image at radius 1, a few seconds at radius 2, and on a 24 megapixel photo at radius 3 it can approach half a minute. The progress bar reflects real work, not an animation. If it is too slow, lower the radius first — the cost rises steeply with it.
Will this make my photo look like plastic?
It can, if you push it. Aggressive noise reduction removes skin pores, fabric weave and foliage texture along with the noise, which is what produces the waxy look. That is why colour-only mode exists and why strength is a blend rather than an on-off switch. It is usually better to leave a little noise than to erase all the texture.
Should I denoise before or after sharpening?
Before, always. Sharpening amplifies whatever is in the image, so sharpening first bakes the noise in permanently and makes it far harder to remove afterwards. Denoise, then sharpen, then resize for output.
Are my photos uploaded to a server?
No. The filter runs on pixels in your own browser using the canvas API. Nothing is sent anywhere, which matters here because noisy photos are often personal ones taken in low light.
Can this remove JPEG compression artefacts?
Partly. The blocky 8 by 8 pattern and the ringing around hard edges are structured rather than random, so a median softens them but does not eliminate them. It helps most on the mottling in flat areas. Nothing recovers the detail that JPEG discarded in the first place.
Does it remove dust and scratches from a scan?
Small dust specks, yes — that is exactly what speckle mode is for, since a dust spot is an isolated pixel far from its neighbours. Long scratches are not isolated, so they survive the filter; those need a clone or heal tool, which is a different kind of operation.
Can I process several images at once?
Yes, and identical settings are applied to all of them. Bear in mind that the cost is per image, so a batch of large photos at radius 3 will take a while. Download individually or take the whole batch as a single ZIP.
Which file formats are supported?
Any format your browser can decode: JPG, PNG, WebP, GIF (first frame), BMP, and AVIF or HEIC where supported. You can keep the original format or export as PNG, JPG or WebP.
Is transparency preserved?
Yes for PNG and WebP output. The alpha channel is copied through untouched and only visible pixels are filtered. JPG has no alpha, so transparent areas are filled with the colour you choose first.
Will the file get smaller after denoising?
Usually, and sometimes substantially. Noise is incompressible by nature, so removing it gives the encoder much more to work with. A noisy photo can drop 20 to 40 percent in JPEG size after a light clean-up, which is a genuine side benefit for web use.
Is this the same as AI denoising?
No, and it is worth being clear about that. Modern AI denoisers are trained to distinguish noise from texture and can reconstruct plausible detail. This is a classical median filter: it is fast, entirely local, needs no model download, and runs privately in your browser — but on very heavy noise a good AI denoiser will produce a better result.
Does the tool add a watermark or need an account?
No to both. No sign-up, no email wall, no daily limit and no watermark. The output is your image with the filter applied and nothing else added.