PNG vs JPG: the answer flips depending on the picture

Neither one is smaller. Which one is smaller depends on what is in the image, and the gap goes both ways by a wide margin. Here is the same 1200 by 900 picture saved both ways, twice, on 2026-09-12:

Same pixel dimensions, same quality setting, opposite answers. Anyone telling you one of these formats is smaller has only tried it on one kind of picture.

Why it flips

The two formats compress by looking for completely different things.

JPG assumes the picture is continuous tone, like the world looks through a lens, where colour shifts gradually and your eye will not miss fine detail it discards. On a photograph that assumption is correct and it pays enormously.

PNG looks for exact repetition instead: runs of identical pixels and patterns it has seen before. A screenshot is mostly flat colour, a logo is a handful of colours, and a chart is large blocks of one value, so PNG finds vast repetition and JPG finds an image full of hard edges, which is the one thing it handles badly.

That is also why a JPG screenshot looks dirty. Sharp black text on white produces faint grey speckle around every letter, which is JPG trying to describe an edge with maths built for gradients. The file is both larger and worse.

The one that is not a trade-off

PNG supports transparency and JPG has no concept of it. If any part of the image needs to be see-through, a logo over a coloured background, a cut-out product shot, an icon, the decision is already made. Saving that as a JPG fills the transparent areas in, usually with black or white, and there is no getting it back.

What ten re-saves of a JPEG actually cost

The usual warning is that every time you open and save a JPG it degrades, and that repeated saves will wreck it. The first half is true. The second is overstated, and it is easy to measure. Taking that photograph and re-saving it at quality 85, comparing each round against the first save:

Out of 255 levels. The damage front-loads into the first save and then very nearly stops, because the same encoder at the same setting keeps making the same decisions about an image that already matches them. The identical test on a PNG moves the worst pixel by 0, which is what lossless means.

Two caveats, because this is the kind of number people over-apply. That is re-saving without editing. Crop, rotate, brighten or resize between saves and the encoder faces a different image each time, so the errors stop converging and start stacking. And a lower quality setting compounds far faster than 85 does.

So which one

JPG for photographs, and for anything going to a person rather than into a workflow: an email attachment, an upload form, a listing photo.

PNG for screenshots, logos, icons, charts, diagrams, anything with text in it, anything with transparency, and anything you will edit again.

WebP if the destination is a website you control. It does both jobs and beats both formats on size, and the only reason it is not the answer here is that plenty of software still will not open one.

filetity is built by Adarsh Mishra. Every number above came from running the conversions on 2026-09-12 with Pillow at quality 85; your images will give different figures, and the direction of the flip will not. Corrections welcome: support@filetity.com.