Knowledge Bank
3 min read
The palette extractor, and reading the share column first
Five colours, their hex codes, and how much of the frame each one actually carries.
Not sure if Package is your weak phase? The free 3-minute workflow audit will tell you.
Channels that look like one channel keep returning to the same five or six colours. Almost nobody writes theirs down. You pick a yellow in the editor, get near enough to last time, drift another shade the upload after, and a year on the shelf looks like six people made it. Every one of those thumbnails looked fine on the day.
Read the share figure before the hex codes
Each swatch comes back with three things: the hex code, the share of the frame it covers, and whether black or white text still reads on top of it.
Run it over your last five thumbnails and you get an honest picture of the palette you actually have. That is usually a different palette from the one you meant to have, and finding the gap takes about two minutes.
What the three readings are for
| Reading | What to do with it |
|---|---|
| Hex code | Paste it into the editor next time rather than eyeballing the shade again |
| Share of frame | Check the colour you think of as yours is actually carrying the image |
| Black or white text | Decide the headline colour before you place the text, rather than after |
That third reading saves the most time in practice. Choosing a text colour by trying both and squinting is a slow loop, and it gets slower once the image has enough going on that both look almost fine.
How the five are picked
Median cut, the quantisation photo editors have used for decades to build a GIF colour table. Every colour in the image starts in one box, the box spanning the widest range gets cut where half its pixels sit either side, and that repeats until five boxes remain.
It is arithmetic on your pixels, so the same image always returns the same palette. Each swatch is a real colour lifted from the frame rather than a suggestion.
Use it on somebody else's thumbnail too
Pull the palette off three videos in your niche that keep winning, and the shared pattern shows up quickly. Not so you can copy it, because copying the dominant look is how you disappear into a feed.
The useful move is the opposite. Once you know what the row around you is built from, you know which colour is missing, and being the only teal thumbnail in a wall of red is a real advantage at no cost.
Nothing is uploaded
The image is read entirely in your browser and never leaves your machine.
The point is the next thumbnail
Extracting a palette is a five-second job and it does nothing on its own. Saving the codes somewhere you will actually look, then opening that file at the start of the next thumbnail, is where the consistency comes from.
A channel that looks deliberate is usually one where somebody wrote five hex codes down once.
Keep reading
Building the image is covered in make the thumbnail, and the composition side belongs to the balance grid. Check the result at real size in shrink-testing the pair before you call it done.