PNG is lossless, so a quality slider has nothing to act on: every pixel is stored exactly, and the only ways to shrink a PNG are to reduce its dimensions or to stop it being a PNG. JPEG is lossy, so dropping quality from 95 to 78 typically halves the file with no visible change. Knowing which kind of format you are holding decides which lever exists.
The right question for a large PNG is almost always whether it should be a PNG at all.
Which format suits which content?
The formats were designed for different pictures, and they behave badly on each other’s.
| Content | Best format | Why |
|---|---|---|
| Photographs | JPEG or WebP | Lossy compression suits continuous tone |
| Screenshots of text | PNG or WebP lossless | Sharp edges survive |
| Logos, flat colour | PNG or SVG | Few colours, hard edges |
| Anything with transparency | PNG or WebP | JPEG cannot store it |
| Web delivery, either kind | WebP | 25–30% smaller at matching quality |
A photograph saved as PNG is the common expensive mistake — it stores every pixel exactly and gets enormous, and converting it to JPEG routinely gives a tenth of the size with no visible difference. A logo saved as JPEG is the opposite mistake, and it produces the fuzzy halo around every edge that JPEG makes on hard transitions.
Why does WebP replace both?
Because it has two modes. Lossy WebP competes with JPEG on photographs at roughly 25 to 30 per cent smaller for matching quality; lossless WebP competes with PNG on graphics and supports transparency in both modes.
Every current browser reads it, which removed the one real objection. What remains is downstream software: some older desktop applications and print workflows still will not open it, which is why converting WebP back to PNG is a common operation despite WebP being the better format.
That conversion recovers nothing. A WebP saved from a website is usually lossy, so the detail was discarded before you received the file — converting to PNG produces a larger file containing exactly the same information.
What does generational loss actually cost?
A little each time, and it accumulates. Every JPEG re-encode discards more, so a file compressed five times looks visibly worse than the same file compressed once at the same setting.
This is the argument for working from an original wherever one exists, and for making format decisions once rather than converting through a chain. A JPEG converted to WebP is a re-encode: the JPEG already threw detail away and WebP now discards a little more of what survived.
Lossless formats have no such problem — a PNG can be opened and saved indefinitely without change, which is why they are the right intermediate for anything still being edited.
What is actually being discarded?
Detail the eye is least sensitive to, chosen by the codec rather than at random. JPEG works in blocks of eight by eight pixels, converts each to a frequency representation, and throws away the high-frequency components first — because the eye notices a change in overall brightness far more than a change in fine texture.
That is why JPEG artefacts look the way they do. Blockiness at low quality is the eight-pixel grid becoming visible, and the coloured fringing around sharp edges is the same process failing on a transition that is entirely high-frequency.
It is also why the format handles photographs well and text badly. A photograph is mostly gentle gradients; a screenshot of text is nothing but hard edges, which is precisely what the algorithm discards first.
Does resolution matter more than quality?
Usually, and by a wide margin. A 4,000-pixel photograph displayed at 800 pixels wide carries twenty-five times more data than it can possibly show.
Capping the longest side is therefore the first move on any oversized image, and the quality slider is the second — which is the order the resize article works through in detail.
The exception is print, where the pixels are being used. There the dimensions are set by the print size and the DPI, and the quality setting is the only lever left.
Questions people ask
Why did my PNG get bigger after compressing it? Because there was nothing to remove and the tool added metadata or a less efficient encoder. PNG is already lossless — a compressor can only reorganise, not discard.
Is quality 75 a safe default? For photographs, generally yes. Below about 70 artefacts appear in smooth gradients — skies and skin first — and above 90 the file grows quickly for a difference nobody sees.
Should I keep the original? Always, if there is any chance of re-editing. Every export is a one-way step, and the original is the only thing that makes a different decision possible later.
Does stripping metadata help? A little, and it also removes location data, camera model and timestamps — which is usually desirable before publishing and occasionally a loss you did not intend.
Is AVIF worth using yet? Smaller again than WebP at matching quality, and supported by every current browser. The remaining gap is the same one WebP had: desktop and print software that has not caught up.
Lossy formats have a quality lever and lossless ones have dimensions. The image compressor, JPG compressor and WebP compressor work on the formats that have something to discard, the PNG compressor is honest about what it cannot do, and PNG to JPG, JPG to PNG, JPG to WebP, PNG to WebP, WebP to PNG and WebP to JPG handle the conversions that actually change the size.