JPG, PNG or WebP: which image format should you choose? Our measurements on two real images

JPG (or JPEG) dates back to 1992. It splits the image into 8 × 8 pixel blocks and discards the fine details the eye notices least. This compression is lossy: the decoded image is never exactly the original, and each new JPG save adds a little degradation. In return, it opens absolutely everywhere.

PNG is lossless: every pixel is restored exactly. It supports transparency, which makes it the natural format for logos, icons and screenshots. Its compression works well on flat colours and sharp text, much less on a photo, where almost every pixel differs from its neighbour.

WebP, introduced by Google in 2010, can do both: lossy compression that is more modern than JPG's, and a lossless mode. It also supports transparency. Every current browser reads it, but some software, government forms and print shops still handle it poorly.

We used two images with opposite profiles. The first is a beach photo taken with a smartphone: 2048 × 1152 pixels, already saved as JPG (529 KB), with a lot of fine texture (pebbles). The second is a photo of a printed letter, as a lossless PNG: 4164 × 9248 pixels, 29.5 MB, mostly black text on a light background.

Each image was converted with FFmpeg using the parameters FileXvert applies at each level: no reduction, balanced reduction and smallest size. Fidelity is measured with the SSIM index, which compares the structure of the converted image with the original: 1 means identical, and above 0.99 the differences are in practice invisible to the naked eye. Below 0.97, flaws become noticeable when zooming in.

On the beach photo, the finding is clear: converting it to PNG multiplies its size by almost eight, without the slightest improvement. The PNG faithfully preserves… the compression artefacts of the original JPG. This is the most common mistake: a PNG never adds quality that was no longer in the source file.

More surprisingly, WebP gains nothing on this image. At a comparable size, the balanced JPG (557 KB) is more faithful than the balanced WebP (573 KB). There are two reasons: the source was already a JPG, whose block structure re-compresses well as JPG, and the very fine pebble texture is a hard case for WebP. Note also that no conversion goes clearly below the original size without visible loss: re-encoding a JPG does not optimise it, it only adds another generation of compression.

With the photographed letter, the story is completely different. The source is a 29.5 MB lossless PNG, far too heavy to send by email or upload to an online form.

Here WebP delivers: at equal fidelity (0.992), it is 23% smaller than JPG; at the balanced level, it is both lighter (1.09 MB versus 1.92 MB) and slightly more faithful. Large areas of plain paper and the sharp edges of text suit it far better than the texture of a nature photo.

Another lesson: going from a 29.5 MB PNG to a JPG or WebP with no reduction already divides the size by six to eight with no visible difference. For a document you need to upload, that is often the only step required.

These two measurements do not replace a test on your own images, but they point to reliable rules:

AVIF, even more recent, often compresses better than WebP. FileXvert cannot produce it today, however: the build of FFmpeg compiled for the browser contains no AV1 encoder, which the format relies on. Rather than offer a conversion that would fail, we removed it from the list. For the web, WebP remains an excellent compromise, read by every current browser.