Removing colour sounds like it should be trivial, and it is the one place where a naive implementation is visibly worse than a correct one. The obvious approach — average the red, green and blue values — produces images that look muddy and wrong in a way most people notice without being able to say why.
Why a flat average fails
Human vision is not equally sensitive to the three primaries. Green contributes far more to perceived brightness than red, and red far more than blue. Averaging them equally therefore misrepresents how bright things actually look: green foliage comes out too dark, blue skies come out too light, and the whole image loses the tonal separation that made it readable in colour.
This tool uses the standard luminance weighting instead — roughly 30% red, 59% green, 11% blue. Those numbers come from measurements of human perception and are the same ones television has used to derive a black-and-white signal from a colour one since the 1950s. The result matches how bright each part of the scene genuinely appears.
What you can expect to change
- Colours that were clearly distinct may become the same shade of grey. A red and a green of similar brightness are very different in colour and nearly identical in luminance.
- Photographs generally gain contrast and structure, since the eye stops being distracted by hue and reads shape and light instead.
- Charts, maps and diagrams often become unreadable, because they rely on colour alone to separate categories.
- Skin tones usually reproduce well, which is why portraiture survives the transition better than most subjects.
It will not make the file smaller
A common assumption, and wrong for this output. The result is still stored as a full-colour image in which every pixel happens to have equal red, green and blue values. PNG compresses that repetition well, so it may shrink a little, but it is not a true single-channel greyscale file and nothing here converts it to one.
If your reason for going grayscale is file size, compressing to a size target will do far more for you — and greyscale genuinely does help there, because a JPEG encoder given an image with no colour information spends almost nothing on the colour channels. Converting here first and then compressing is a reasonable sequence for hitting a very small target.
Transparency is preserved
Only the colour channels are altered; the alpha channel is untouched. A logo with a transparent background stays transparent, which makes this usable for producing a monochrome version of an asset for use on coloured backgrounds. Everything runs in this tab with nothing uploaded.