Joining pictures that aren’t the same size
This is the commonest reason a join looks wrong, and it is settled in one step. Here is what each of the four answers actually does to your pictures.
The problem, stated plainly
Pictures joined in a row share one measurement: their height. Joined in a column they share their width. If the pictures you have are not already the same on that shared axis, something has to give — either a picture changes size, or a band of empty space appears beside the smaller one. There is no third possibility, and a tool that pretends otherwise is just picking one for you without saying so.
So the step asks. Four answers, and the interesting part is what each one costs.
The four answers, and their price
Match the biggest
Everything grows until it meets the tallest picture in the set. Nothing is thrown away, and the result is as detailed as the best picture you put in. The cost falls entirely on the small ones: a 600-pixel-tall picture stretched to 2,400 has four times as many pixels as it has information, and they are all guesses. On a photograph of a landscape that is barely visible. On a screenshot with text in it, it is very visible indeed.
Match the smallest
Everything shrinks to the shortest. Every picture stays as sharp as it was or gets sharper, because downscaling throws detail away rather than inventing it. The cost is that your best picture is no longer your best picture, and if the shortest one is much shorter than the rest you have thrown away a great deal to match it.
Match the first
Whichever picture sits first in the tray sets the rule and the others follow. This is the right answer surprisingly often: when one picture in the set is the subject and the rest are supporting, the subject should be the one that keeps its size.
Leave them alone
Nothing is scaled. Each picture is placed at exactly the size it arrived, and the gaps beside the shorter ones are filled with whatever ground colour you chose. This is the only answer that guarantees no picture is altered at all, and it is the one people reach for last because a band of colour looks like a mistake.
When padding beats scaling
Three cases where the empty band is the better result:
- Anything with text in it. A screenshot, a receipt, a scan of a form. Resampling text is the operation people notice first — letters go soft, thin strokes break up, and the result reads as low-quality even to someone who could not say why. A white band beside a crisp screenshot reads as a layout choice.
- Documentation and evidence. If the picture is proof of something, altering its pixels at all is worth avoiding. Padding leaves both files pixel-identical to what you were given.
- Sets where one picture is wildly different. Five pictures at 3,000 pixels and one at 400: matching the smallest destroys the five, matching the biggest destroys the one. Padding leaves the five alone and the odd one out looks like the odd one out, which is honest.
A portrait joined to a landscape
This is the case that surprises people. Put a tall picture next to a wide one in a row and match the tallest, and the wide picture becomes enormously wide — a 4,000 × 3,000 landscape matched to a 3,000 × 4,000 portrait grows to 5,333 pixels across. The finished picture is over 8,000 pixels wide and mostly landscape. Nothing has gone wrong; the arithmetic simply has consequences that are hard to picture in advance, which is why the size readout under the preview updates as you change the setting rather than waiting until you save.
For a portrait and a landscape specifically, two things usually work better than a row. Stack them in a column instead, where the shared measurement is width and the mismatch is far gentler. Or scale nothing and centre them, which gives you both pictures untouched with a band of colour under the wide one.
How the scaling is done
When a picture does have to change size, it steps down in halves with a careful final pass rather than jumping straight to the target in one go. That sounds like a detail and it is the difference between text that stays legible and text that turns to mush: a single large downscale samples too few of the original pixels and drops whole strokes of a letter between them. Halving repeatedly keeps every pixel contributing to the result.
None of this recovers detail that was never there. A picture made bigger to match its neighbours will always be softer than one that was that size to begin with, however carefully it is done, and no setting on this page changes that. If a small picture matters more than the join does, the honest move is to scale nothing.