Guide
Fit, cover and stretch, on one photograph
Almost nobody knows which of the three they want until they have seen what each one does to their own picture. Here is what each one does.
The problem all three are solving
You have a picture with one shape and a box with another. A 3:2 photograph from a camera, say, and a request for 800 × 600, which is 4:3. Those two shapes cannot both be satisfied. Something has to give, and there are exactly three things that can give: the dimensions you asked for, some of the picture, or the proportions of everything in it.
That is the entire subject. Fit gives up the dimensions. Cover gives up part of the picture. Stretch gives up the proportions. Every resizer in the world offers some version of these three, usually under different names and often without saying which is which.
Fit: 800 × 533
Fit scales the picture until it sits entirely inside the box. Every part of the original is in the result, nothing is added, and the result is not the size you typed — one of your two numbers is met and the other one is smaller than you asked for.
From a 3000 × 2000 photograph into an 800 × 600 box, fit produces 800 × 533. The width is exact. The height is 67 pixels short, because a 3:2 picture 800 pixels wide is 533 pixels tall and there is no arrangement of arithmetic that makes it 600.
Fit is the right answer far more often than people expect. If the requirement is really a ceiling — no wider than 800, no taller than 600, keep it looking like the original — fit is what that requirement means, and the fact that the height comes out at 533 is not a failure.
Cover: 800 × 600, minus a strip
Cover scales the picture until it covers the box completely, then discards whatever hangs over the edges. The result is exactly the dimensions you typed. The cost is that some of your picture is not in it.
From the same 3000 × 2000 photograph, cover reads a 2667 × 2000 rectangle out of the middle and resamples that to 800 × 600. A 333-pixel column has gone — 167 from each side with the anchor at centre, or all 333 from one side if you moved the anchor.
The nine-point grid is the whole of the control here. It is nine positions rather than a rectangle you can drag, and that boundary is deliberate: choosing the rectangle yourself is cropping, and cropping wants its own screen, its own handles and its own set of ratio presets. Nine positions covers the cases where the answer is obvious — the middle, or the top, or a corner — and does it as arithmetic.
Stretch: 800 × 600, and everyone is wider
Stretch scales the two axes by different amounts until the picture is exactly the box. Nothing is discarded and the dimensions are exactly what you typed. Everything in the picture is now the wrong shape.
From 3000 × 2000 to 800 × 600, the horizontal axis is scaled by 0.267 and the vertical by 0.3. That is a 12% horizontal squash: circles are ellipses, faces are narrower, and text in the picture leans. It is visible on a photograph of a person immediately, and much harder to see on a landscape, which is why people ship it by accident.
It is offered anyway, and labelled with what it will do. Some specifications genuinely demand exact dimensions from a source of the wrong shape and forbid cropping; some source material is a texture or a gradient where proportion does not mean anything. Hiding the option does not stop anybody needing it — it just sends them to a worse tool.
The chain switches the question off
When the aspect chain is on, the second field is derived from the first using the shape of your source, so the box you are asking for already has the picture's shape. All three modes then produce identical pixels, and the mode buttons switch themselves off to say so.
There is no fourth mode here that pads the picture out to the box with bars. Adding pixels that were never photographed is a different operation with its own decisions — what colour, how thick, on which edges — and this page only ever resamples what is already in the frame.