Does removing metadata change the image itself?

The two approaches

Metadata sits in sections of the file that are separate from the compressed picture. A tool can therefore either rewrite just those sections, or decode the whole image and encode it again.

The first is lossless: the compressed data is copied byte for byte and never decoded, so the output produces pixel-for-pixel identical results. The second loses quality every time, because JPEG compression is lossy and re-applying it compounds the damage.

How to tell which you are using

A container rewrite is fast and produces a file only slightly smaller than the original — smaller by roughly the size of the metadata that was removed.

A re-encode is slower, and the output size will often change substantially in either direction as the encoder makes its own choices. If a tool offers you a "quality" slider while removing metadata, it is re-encoding.

What this tool does

Container rewriting only, for every supported format. The test suite asserts it directly: it extracts the JPEG scan data and the PNG IDAT chunks before and after cleaning and requires them to be byte-identical.

It also cleans a file twice and requires the two outputs to match exactly, which proves the operation is idempotent — there is no generational loss to accumulate.

What can change appearance

Two pieces of metadata affect how an image is displayed, even though neither is part of the pixel data.

• The orientation flag tells a viewer to rotate the image. Remove it and a photo taken sideways displays sideways — the pixels are unchanged, but what you see is not.
• The ICC colour profile tells colour-managed software how to interpret the colours. Remove it and colours can shift slightly in applications that were using it.

Which is why there are two modes

A quick clean keeps those two and removes everything else, so the picture looks exactly as it did. A full clean removes them too, for when you want nothing left at all and can accept rotating a photo afterwards.

The before-and-after report lists anything kept on purpose separately from anything that survived unexpectedly, so the distinction is never hidden from you.

The same holds for video

Video is where re-encoding would hurt most, and where it is least necessary. The metadata is in the container, not in the encoded stream, so there is no reason to decode a single frame.

Cleaning a video here rewrites the box structure and copies the mdat payload — every encoded frame — through untouched. A multi-gigabyte file finishes in seconds, and the output is bit-for-bit identical media.

Track rotation is deliberately preserved, for the same reason the orientation flag is kept in a quick clean: removing it would make a video shot on a phone play sideways.

How to check for yourself

You do not have to take this on trust. Clean a JPEG, then compare the output with the original in any tool that can show file bytes: everything from the start-of-scan marker onwards will be identical.

A simpler check is to clean the same file twice and compare the two outputs. If they match exactly, nothing is being re-encoded — a lossy pipeline would produce a different result each time.

Questions