ConvertPicturesConvertConvert a photo

PNG vs JPG, measured: which one each picture needs

Not a rule of thumb: three kinds of picture written both ways at matched visual quality, with the byte counts that came out.

The measurement

Each picture below was dropped into this converter with quality on Visually identical. PNG is encoded once (it is lossless); JPEG is searched to the lowest quality that still measures 0.990 or better against the source at 512 px, and on the larger pictures its busiest 512 px crop is also scored at full resolution. WebP is shown too, because on two of the three it is the answer neither PNG nor JPG is.

PicturePNGJPG (searched)WebPRecommended
Harbour at eveningphotograph-like scene with grain, 1280×8601425 KB90 KBq 73 · 0.9905 at 512 px · crop 0.8899175 KBlossy, matchedJPG — a sixteenth of the PNG
Dashboard screenshotflat interface with text, 1100×70058 KB25 KBq 31 · 0.9961 at 512 px · crop 0.979511 KBlosslessLossless WebP; of the two, PNG
Ring logodrawn, on transparency, 640×64030 KB9.3 KBq 31 · 0.9926 at 512 px8.6 KBlosslessLossless WebP or PNG — JPG loses the transparency
Measured on 2026-10-23 in headless Chromium 153 on Linux, on this converter’s own sample set (the pictures “Try a sample set” draws). Byte counts depend on the browser’s encoders; run it on your own pictures for your own numbers.

What the numbers say

On the photograph-like scene the matched JPEG is a sixteenth of the PNG — 90 KB against 1.39 MB. Grain and gradients defeat PNG’s prediction, and JPEG discards detail the eye does not track. Lossy WebP, held to the same bar, came out larger than the JPEG on this picture: the “WebP is a quarter smaller” rule of thumb is an average, not a promise.

On the screenshot, the matched JPEG is smaller than the PNG in bytes — but look at how it got there. The 512 px score is generous to JPEG on text, so the search went all the way down to quality 31; the full-resolution check of the busiest crop then measured 0.9795, below the 0.990 bar, and the card says so. That is the ringing around letters that makes JPEG the wrong tool for a screenshot. Lossless WebP was smaller than both, with every pixel exact.

On the logo, JPEG is not a candidate at all: it cannot hold the transparent background, so its byte count is for a version flattened onto white. Lossless WebP keeps the transparency at under a third of the PNG.

So the question is not which format is better but what the picture is. Photographs want lossy; drawn pictures want lossless; transparency rules JPEG out. This converter classifies each file from two measurements and prints the rule that decided — and when a route fixes the target, it writes what you asked for and says beside it when the measurement disagrees.

Questions

Is PNG better quality than JPG?
PNG is lossless, so it never loses detail; JPEG is lossy. But at a quality searched to measure visually identical, a JPEG of a photograph looks the same as the PNG at a fraction of the size. Quality only favours PNG on drawn pictures, where JPEG rings around edges.
Should a screenshot be PNG or JPG?
PNG, or lossless WebP. A JPEG can come out smaller in bytes, but flat colour and sharp text are where it is worst: letters soften and ring, which a full-resolution check catches even when a downscaled score does not.
Should a photo be PNG or JPG?
JPG. A photograph's grain defeats PNG's prediction, so the PNG is many times larger for no visible benefit.
What about a logo with a transparent background?
PNG or WebP. JPEG has no transparency at all, so the background would have to become a solid colour.
Does converting JPG to PNG improve it?
No. The PNG keeps the JPEG's pixels, artefacts included. It only stops further loss if the picture will be edited and saved again.