Image Converter

The canvas in the middle

A conversion is three steps, and the middle one is where the losses live. The file is decoded into raw pixels, those pixels are drawn onto a canvas, and the canvas is encoded into the target format. Every format in the list can be decoded into that middle state and every one can be written out of it, which is what makes conversion between any two of them possible — and the middle state is the lowest common denominator of the whole set.

Eight bits per channel, because that is what a canvas holds. GIF and 256-colour PNG arrive indexed and leave as full truecolour, so a small palette image can come back several times larger even though nothing about it was lossy. Animation does not survive at all, because a canvas has one frame; an animated GIF converts to a still of its first frame. An embedded colour profile is dropped unless it is sRGB, so a wide-gamut original can come back flatter. Sixteen-bit files lose their extra precision. None of this appears in the stat line, and all of it is worth knowing before the original is thrown away.

Alpha is a channel, not a colour

Formats that carry transparency store a fourth value per pixel saying how much of whatever is underneath shows through. Formats that do not have no way of writing that value down, so a transparent image moving into one has to be flattened against something — and the choice of that something is visible. Black produces the familiar dark box behind a logo; white makes a pale logo vanish on a pale page. The matte control exists to pick the colour the image is actually going to sit on.

Flattening costs more than the background, too. The edge of a shape drawn on a transparent canvas is not solid: each boundary pixel is a blend of the shape and whatever was behind it, which is what keeps the outline from looking like a staircase. Composited onto a new colour, those blended pixels keep the old background mixed into them, and the result is a light or dark fringe tracing the whole silhouette. It is most obvious on text and thin line art, and it is the reason a flattened logo sometimes looks chewed along its edges rather than cleanly cut.

Where the format stops mattering

Format and quality are not competing for the same bytes, and knowing which one is doing the work tells you which control to reach for. Container overhead is a fixed cost: a JPEG carries its header, its quantisation tables and often a thumbnail; a PNG carries chunks; WebP carries a RIFF-style container. At small dimensions that overhead can be most of the file, which is why converting a 64-pixel icon from PNG to JPEG can make it larger:

64 × 64 icon        ~500 bytes of container overhead
                          ~200 bytes of actual image data
        -> the wrapper is the file. Format choice dominates.

      1600 × 1200 photo   ~700 bytes of container overhead
                          ~400 KB of image data
        -> the wrapper is noise. Format and quality dominate.

Below roughly ten kilobytes the difference between the three formats stops being worth a decision — the bytes saved are smaller than the variation between two encodes of the same file. Above a couple of hundred kilobytes it is the largest lever on the page, and a move from JPEG to WebP takes off a third of the weight without the softening that comes from leaning harder on a lossy encoder. The quality slider decides how much quality you are willing to trade. The format decides whether the trade was necessary at all.

Converting in the direction that helps

Direction matters as much as the target. A photograph saved as PNG lands several times larger than the JPEG it came from, because PNG writes every pixel down exactly and the JPEG did not. A JPEG turned into WebP gets a smaller file containing the same damage — the compression happened when the JPEG was written, the artefacts are part of the pixels now, and no container takes them back out. That is why the useful direction is always from the original: an uncompressed master exported once into whatever the destination wants, rather than a chain of conversions hoping the last one cancels the earlier ones.

If a file needs to be a different size as well as a different format, change the size first — a shrunken image has less detail to encode, so the format decision is made on fewer pixels and the result is smaller for both reasons. The Image Resizer handles that half, and the Image Compressor is the one to reach for when the format is already right and only the bytes need attention.