Dots or instructions

Every image is stored one of two ways, and that choice decides whether you can enlarge it, whether you can edit it later, and whether the AI can touch it.

Bitmap (or raster): a grid, and a very long list of numbers — one per cell, saying what color it is. The photo you take with your phone is twelve million cells and nothing else: the file does not know there is a face, a table or a sign in there. Only colors in positions.

Vector: it does not store cells, it stores instructions, and it stores them written in letters. "A line from here to here. Red fill. A text that says HELLO, this size." The drawing is rebuilt every time you look at it, at whatever size is needed.

You have lived this. You downloaded a client's logo from Google Images, pasted it into the slide deck, stretched it to fill the slide — and it came out blurry, with staircase edges.

You did nothing wrong: you asked a bitmap for more information than it had stored.

It is the same thing that happens when you zoom into a screenshot until the letters turn into smudges.

A close cousin: the photo forwarded fifteen times. The one that arrives wrecked suffers something similar but different: nobody enlarged it — what happens is that every forward compresses it again and throws away a little more information to make it lighter.

The result looks alike because the root is the same: a bitmap can only lose. It never recovers.

A vector does not have that problem, because there is nothing to compress: it is four instructions that take up what they take up.