What a converter keeps, strips, and quietly rewrites

Every image carries a second file inside it: where it was taken, which camera, which colour space, which way is up. Conversion treats each of those differently, and only one of them is guesswork.

Open a photo and you see pixels. The file also contains a block of metadata describing them, and when you convert the file, that block does not travel automatically. Some of it is dropped on purpose, some is applied to the pixels and then discarded, and some is preserved because losing it would visibly change the image. Knowing which is which explains most of the small surprises people report after a conversion.

EXIF: dropped, and that is usually the point

EXIF is the camera's notebook: model, lens, exposure, the timestamp, and on a phone the GPS coordinates of where you stood. It is written by the camera and read by photo libraries, and it is the reason a screenshot of a house can also be a record of an address.

Conversions here do not carry EXIF into the output. That is a deliberate default rather than an oversight: the most common reason to convert a photo is to give it to somebody, and handing over coordinates you did not mean to share is a worse failure than losing an exposure setting. If you need the metadata, read it from the original before converting, and keep the original. It remains the only copy that has it.

Orientation: applied, then removed

Phone sensors do not rotate. When you hold a phone sideways the pixels are stored sideways and a single EXIF tag records how the viewer should turn them. This one tag is responsible for an enormous share of all image confusion: the same file looks upright in one app and rotated in another, because one app read the tag and the other did not.

Colour profiles: kept, because the alternative is visible

A pixel value of pure red does not name a colour on its own. It names a coordinate, and the colour profile says which space those coordinates belong to. Recent iPhones capture in Display P3, a noticeably wider space than the sRGB the web assumes. The same numbers, interpreted as sRGB, come out duller and shifted, most obviously in saturated reds, sunsets and skin tones.

So the embedded profile is carried into the output wherever the destination format can hold one. This matters most on HEIC to JPG and HEIC to PNG, where the source is almost always a phone photo in P3. If colours still shift after a conversion, the usual culprit is downstream software that ignores profiles entirely, which is common in older document tools and some print workflows. Converting to sRGB deliberately, before handing a file to that kind of tool, is the fix.

Transparency and animation: format-limited, not policy

Some losses are the format's, not the converter's. JPG has no alpha channel, so a transparent PNG converted to JPG gains a solid background, white unless something else is specified. This is why a logo that looked fine on a dark page suddenly sits in a white box, and it is the one thing to check on every PNG to JPG conversion. PNG and WebP both keep alpha, so PNG to WebP is the safer move for anything with a cut-out edge.

Similarly, converting an animated file to a still format keeps one frame. Nothing is broken; there is simply nowhere for the other frames to go.

Documents carry more than you can see

A Word file's metadata is more sensitive than a photo's. It can include the author name, the total editing time, the template it came from, and, if markup is displayed, text that someone believed they had deleted. Converting DOCX to PDF prints what is currently shown, so tracked changes and comments become permanent, readable parts of the output. We measured that one across twelve generated documents, and deleted text really does arrive in the PDF.

Before converting anything that leaves your organisation: accept or reject all changes, delete comments, and check the document properties. A PDF is a photograph of the document, and photographs record what was in the frame.

Caveats

• This describes the behaviour of this site's pipeline. Other converters make different choices, particularly about EXIF: several preserve it by default, which is the opposite of the trade made here.
• Profile preservation is bounded by the destination format. Where a format cannot carry an ICC profile there is nothing to preserve, and the pixels are what you get.