Convertir imagen

Converting an image is a decode followed by an encode. The source is turned into raw pixels, and those pixels are written out under a different set of rules. Nothing is translated directly between formats, which is why a conversion is never entirely free and why the choice of target matters.

The transparency trap

JPEG has no alpha channel. When a PNG with a transparent background is converted to JPEG, those pixels have to become something, and a canvas that has not been painted is transparent black — so they become black. This is behind almost every “my logo came out on a black square” report on the internet.

The fix is one rectangle: fill the canvas with a chosen colour before drawing the image, so semi-transparent pixels composite against it properly instead of being interpreted as zero. That colour is a control here rather than a hard-coded white, because a logo built for a dark site needs a dark background and white would be just as wrong as black.

What each format is for

JPEG is lossy, universal, and has no transparency. It is built for continuous tone, which is why it is excellent on photographs and poor on text — its artefacts hide in detail and stand out against flat colour. Every device made in the last thirty years opens it, which is still the reason to choose it.

PNG is lossless and keeps transparency. Screenshots, logos, diagrams and anything with sharp edges belong here; a lossy format smears exactly the edges that make them readable. The cost is size: a photograph as PNG is often ten times the JPEG.

WebP is typically 25–35% smaller than JPEG at the same visible quality, supports transparency, and has a lossless mode that usually beats PNG. It is supported by every browser in current use and is the sensible default for the web now.

AVIF is smaller again — often 20–30% below WebP — and noticeably better at low bit rates, where JPEG falls apart. The costs are encoding time and availability: not every browser can produce one. This page tests that by encoding a 1 × 1 image before offering the option, rather than guessing from the user agent, which would be wrong within a year.

Lossy to lossless does not undo the loss

Converting a JPEG to PNG produces a lossless copy of an image that has already lost information. The artefacts are now preserved perfectly and the file is several times larger. It is occasionally the right move — if the image is about to be edited repeatedly, a lossless intermediate stops the damage compounding — but it is not a way to recover quality, and nothing is.

Going the other way costs a generation

Converting a JPEG to another JPEG, or to WebP, re-quantises data that was already quantised. Each pass adds artefacts on a block grid that may not line up with the previous one. Convert from the original whenever you have it.

What happens to metadata

Because the pipeline goes through raw pixels, EXIF blocks do not survive: camera model, timestamp and GPS coordinates are all dropped. For most uses that is a feature — a photo posted online carries its location otherwise. The single tag that is honoured rather than discarded is orientation, which is applied to the pixels during decoding so that a sideways-shot photo comes out upright.