color palettes in design
Extracting a palette from an image is not about picking the "prettiest" colors: it is about grouping thousands of pixels into a handful of clusters and choosing one representative color per cluster. This tool uses median-cut quantization, the same classic image color-reduction algorithm, with a specific tweak so it never returns a tone that does not actually exist in the source photo.
A median-index cut would leave the blue entirely on the red side, and the palette would lose that color.
The plain average of those pixels would be a gray that does not exist at any real point in the photo.
Only the icon's opaque art enters the count that produces the palette.
The tool uses the median-cut algorithm: it repeatedly splits the image's color space along the axes of greatest variation and then, inside each group, returns the most frequent tone (the mode) rather than the average. Averaging would produce washed-out greys that do not exist in the image. Each color comes with the share of the image it covers.
Because the algorithm answers "which colors dominate this image by pixel count", not "which colors look good together"; a sunset photo returns several close shades of orange, because that is genuinely what dominates the image, not a palette curated by a human.
Up to 16 swatches per extraction; the median-cut algorithm splits pixel boxes until it reaches the requested count or that cap, whichever comes first.
Because the modal color is computed over a 4096-bucket grid (4 bits per channel): a pixel that falls just past a bucket boundary changes which bucket is the most frequent in the box, and the final color shifts along with it.
No: pixels with an alpha value below 16 are discarded before any calculation, so only the visible pixels of the image contribute to the extracted colors.
Color analysis happens entirely in your browser. No image is sent to any server.