Denoise
A median filter is the right tool for speckle. Where a blur averages a bad pixel into its neighbours and spreads the damage, a median sorts the neighbourhood and takes the middle value, so an outlier is discarded outright instead of smeared. That is why it removes sensor noise and JPEG artefacts while leaving edges comparatively intact.
Drop images here
PNG, JPG, WebP — multiple files allowed
Radius and strength are different dials
Radius decides how far the filter looks. A radius of 1 examines the nine pixels around each pixel and is enough for light grain and mild JPEG blocking; radius 3 examines forty-nine and will flatten heavy noise, along with any texture finer than about seven pixels. Skin pores, fabric weave and fine text all live in that range, so a large radius has a plastic look that is hard to undo.
Strength is a straight blend between the original and the filtered result. Running radius 2 at 50% often beats radius 1 at 100%: you get the outlier rejection of the bigger window but keep half the original detail. That combination is the usual sweet spot for a noisy phone photo. The cost of a median is real, though — it sorts a window per channel per pixel, so radius 3 on a 12-megapixel image takes a few seconds.
The chroma-only mode
Noise is two problems wearing one coat. Luminance noise — grain, brightness varying pixel to pixel — reads as texture and is often not objectionable at all; film had plenty of it. Chroma noise, the coloured confetti that shows up in shadows at high ISO, always looks like a fault.
Ticking "colour noise only" filters the image and then restores the original brightness of every pixel, keeping only the colour change. The result is that the confetti goes and the detail stays, because detail is carried almost entirely by luminance. It is the first thing to try on a dark indoor photo, and it lets you use a larger radius than you could otherwise get away with.
FAQ
Will this fix a blurry photo?
No — it removes noise, and noise and blur are opposite problems. A median filter can only make a soft image softer. Sharpen is the other direction, and the honest answer is that detail which was never captured cannot be recovered.
What about salt-and-pepper noise from a bad scan?
That is the case a median filter was invented for and it handles it almost perfectly. Isolated black or white pixels are pure outliers, so even radius 1 at full strength removes them without touching anything else.
Is it better to denoise before or after resizing?
Before, if you are downscaling significantly — otherwise you throw away detail while keeping the noise that survived the averaging. If you are upscaling, denoise first as well, because upscaling magnifies noise faster than it magnifies detail.
Does it touch transparency?
No. The alpha channel is passed through unchanged, so a cut-out object keeps exactly the edge it had.