The colors in any photo.

No. The photo is drawn into a small canvas in your browser and the clustering runs in JavaScript on that. Your photo never touches our servers — we don't run an image pipeline at all.

The photo is sampled down to 260 pixels wide, every sample is converted to OkLab, and k-means with k-means++ seeding finds the colours that best explain the whole image. OkLab matters because distances in it track what an eye sees, so the swatches come back genuinely different from each other rather than three near-identical greys. The seeding is deterministic, so the same photo always gives the same palette.

Sometimes because the photo really is five shades of beige. More often it is because area and interest are different questions: a wall can cover half the frame while the colour you actually remember covers a corner. That is what the bias slider is for — left picks the colours that cover the most of the photo, right picks the ones you would name if asked to describe it. The largest colour stays in the palette either way.

Yes — up to 8 on the colours slider. Past that a palette stops reading as a palette; if you need twelve, run the photo twice at different bias settings and merge the results by hand.

Click a chip; the label says 'copied' for a moment. The panel on the right also has the whole palette as a comma-separated list, as CSS custom properties, and as a JavaScript array, each with its own copy button. If your browser refuses clipboard access the tool says so rather than pretending it worked.

A 1400×680 palette card with the chips, hex codes and the share of the frame each colour accounts for, on a paper-cream ground. Ready to drop onto a Pinterest pin, an Instagram caption, a Notion page or a deck.