Processed here, not somewhere else
The image is decoded, drawn through a filter chain into a canvas, and encoded again — three local steps and no network involvement. Open the network tab and it stays empty from the moment you choose a file. Disconnect and the tool keeps working, which is the only demonstration of this claim worth anything.
What each control does to a pixel
Greyscale collapses red, green and blue into one luminance value weighted 21% / 72% / 7%, matching how the eye responds. A plain average would make greens too dark and blues too light. Sepia applies a fixed colour matrix on top, warming the result the way a toned print does.
Brightness multiplies every channel by the same factor, so it lifts the whole tone curve and clips at the top. Contrast pushes values away from mid-grey (128) or pulls them toward it, expanding the range at the cost of the extremes. Saturation keeps each pixel's luminance and scales only the distance from it, so 0 gives the same result as greyscale and values above 100 push colours further apart. Invert subtracts each channel from 255. Blur is a gaussian convolution with the radius you enter.
Order matters more than people expect
The chain runs greyscale, sepia, saturation, brightness, contrast, invert, blur. That is fixed, and knowing it explains results that otherwise look wrong.
Because brightness runs before contrast, raising both means the highlights clip first and the contrast pass then stretches what is left of a shortened range — which is how a rescue attempt turns into a blown-out photograph. Go in small steps: brightness 110 to 120, contrast 105 to 115, look, adjust. Saturation is meaningless once greyscale is on, since there is nothing left to saturate. Sepia after greyscale wins visually, so ticking both looks like sepia alone.
Practical starting points
| Problem | Try |
|---|---|
| Photo shot in poor indoor light | Brightness 115, contrast 108 |
| Flat, hazy outdoor shot | Contrast 115, saturation 110 |
| Black and white for a printed form | Greyscale on, contrast 110 |
| Muted background for text over it | Brightness 115, saturation 70, blur 3 |
| Scanned document that looks grey | Contrast 130, brightness 105 |
Beyond about 150 on brightness or contrast the tonal steps become visible as banding, particularly in skies and skin, and JPEG compression exaggerates it. If you need that much, the exposure was wrong at capture and no amount of arithmetic will invent the detail that was never recorded.
Blur is not concealment
It is worth separating the two uses. As an aesthetic treatment — softening a background so text sits on top of it — blur is exactly right. As a way of hiding a face, a licence plate or a line of text, it is unreliable, because blurring spreads information rather than removing it, and a modest blur over a known font or a limited character set has been reversed in practice. If something must be unreadable, use the pixelate tool at a heavy setting, or cover it with a solid block. Do not use a blur and assume you are done.
What the export costs you
Every pass through this tool re-encodes the file. On JPEG that is another generation of compression loss on top of whatever the source already had, so keep the quality at 90 or above and work from the original rather than iterating on your own output. Choosing PNG makes the round trip lossless at the cost of a much larger file for a photograph. And as with every canvas tool, the EXIF block and any embedded colour profile do not survive — check the original with the EXIF viewer first if you need to know what you are discarding.
Questions people ask
Can I filter only part of the image?
No — everything here applies to the whole frame. Selective adjustment needs a mask, and a mask needs a drawing surface with brushes and feathering, which is the definition of an image editor rather than a form. The nearest thing available here is the region control in the pixelate tool, which affects a rectangle or an ellipse you specify by percentage.
Where are white balance and curves?
Not implemented. Brightness, contrast and saturation are the three that can be described honestly as a single number each. Colour temperature needs a per-channel adjustment, and curves need a control surface. Both belong in a raw processor, where you would also get the benefit of working on the original sensor data instead of an already-compressed JPEG.
The result looks softer than the original.
Check the blur value is zero first — it is easy to leave a number in that field. If it is zero, what you are seeing is JPEG re-compression. Raise the quality to 95 or set the output format to PNG and compare; if the PNG looks identical to the source, compression was the cause and the filter chain was not.
Why does the greyscale look different from the black and white mode on my phone?
Phone modes usually apply a tone curve alongside the desaturation, and often a contrast boost, because a straight luminance conversion tends to look flat. To get closer, turn greyscale on and add contrast around 110 to 120. There is no single correct black and white conversion — it is an interpretation, and different software interprets differently on purpose.