Find the dominant colours
Extract the dominant colours from an image.
Working…
Questions people ask
- How are the colours extracted?
- The image is oriented, sampled on a fixed grid of at most 4096 pixels, and then median-cut clustered: the box with the widest colour spread is split at the median repeatedly until the requested palette size is reached, and every cluster is named by the median of its member pixels. The same image always produces the same palette.
- What does the share number mean?
- The cluster's slice of the sampled grid, to one decimal. It is an estimate, not an exact pixel frequency of your original: the sample is up to 4096 pixels and shares are rounded, so they may not total exactly 100%.
- How are transparent pixels treated?
- Fully transparent pixels (alpha below 8 after orientation) are excluded from sampling and counted in the report. An image that is entirely transparent is refused with that fact stated. Semi-transparent pixels keep their stored RGB values: they are not blended with white and alpha is not premultiplied.
- What is the swatch strip?
- A fixed-width PNG that paints one horizontal band per palette colour, in palette order, so you can see the whole palette at a glance or drop it into documentation. Both artefacts are byte-for-byte reproducible: the same input always gives the same files.
- Will JPEG colour noise fill my palette?
- Clustering absorbs it: the median of a cluster sits inside the dominant tone even when neighbouring pixels vary. Extremely noisy or dithered images may still produce near-duplicate palette entries at 10 colours; 5 or 8 usually reads better.
- What happens to my image?
- The upload is deleted as soon as processing finishes; the palette report and the swatch strip stay downloadable for about 10 minutes, then are deleted automatically.