// what it does and why
Drop an image onto the page and it gives you the six colours that image is mostly made of. Click any one of them and the hex code is on your clipboard. The picture never leaves your device.
Open it →Pick a file, or drag one anywhere onto the page. A small preview appears, and beside it six swatches, ordered so the colour covering the most of the image sits at the top and the rarest of the six sits at the bottom. Click a swatch and its label swaps to a short confirmation so you know the copy landed, then goes back.
That is the whole tool. No account, no export, no project to save. It answers one question and gets out of the way.
I kept doing this by hand. Screenshot a piece of art, open an image editor, eyedropper one pixel, write down the hex, eyedropper another, and afterwards still not know whether the colours I picked were the ones carrying the image or just the ones my cursor happened to land on.
The other reason is that the obvious alternative is uploading the picture to a site you have never heard of. For somebody else's artwork, or a photo of a person, that is a bad trade for six hex codes. This version cannot leak anything because it never sends anything. The file is read straight into the page and the work happens on your own machine, offline if you like.
It scatters six markers at random through the range of possible colours, then sorts every sampled pixel to whichever marker is nearest. Each marker moves to the average of everything it caught, and the sorting runs again. After a handful of rounds the markers settle into the six clumps the image really has, and the number of pixels each one caught decides the order on screen. The method is called k-means clustering. It is old, small, and good at exactly this.
Two things follow from that. The colours you get back are averages of a group of pixels, so a hex code it hands you might not appear anywhere in the original image. And because the markers start in random places, running the same image twice can give you slightly different sixes. Usually that is a shade of difference. On an image with no clear groups in it, sometimes more.
It reads the preview, not your original. Everything is scaled down so the longest side is 300 pixels before a single colour gets counted, which is why the answer arrives instantly even on a phone. The cost is real: a small bright accent, a logo in a corner, a thin band of sky, can be averaged into its neighbours during that resize and never make the list.
Six is fixed, with no slider. Fully transparent pixels are skipped, so a logo saved on a transparent background gives you the logo's own colours rather than a palette of whatever sits behind it. And nothing forces the six apart, so an image that genuinely is mostly one colour will honestly return several near duplicates of it.