Get Colors Image

It uses a technique called median-cut quantization: it treats every pixel's color as a point in 3D red/green/blue space, repeatedly splits the most populous group of colors in half along whichever color channel varies the most within it, and once it has as many groups as colors you requested, averages each group into one representative color. Groups with more pixels come first in the results, so the palette reflects how much of the image each color actually covers.

Averaging everything at once would blend a photo with distinct red and blue areas into a muddy purple that appears nowhere in the original. By splitting the color space into separate groups before averaging, each group only contains similar colors, so red areas and blue areas stay represented as their own separate, recognizable palette entries instead of canceling each other out.

If that color only covers a small area of the image, it may end up folded into a larger, more dominant group rather than getting its own slot, especially at lower color counts. Increasing the number of colors requested gives small or subtle color areas a better chance of being split out into their own palette entry.

The image is first downscaled to a smaller working copy before analysis, since palette extraction only needs a representative sample of colors, not every pixel, and this keeps the tool fast even on large photos. This does not affect the image itself, only the analysis, nothing is uploaded and the original file is never modified.

Yes, click "Copy as CSS Variables" to copy the palette as ready-to-paste --color-1 through --color-N custom properties, or "Copy as HEX List" for a plain list of hex codes.