When JPEG is the right target
Really one case: the source is a photograph that ended up as a PNG. That happens constantly — a screenshot of a photo, an export from a design tool, a download that defaulted to PNG. The file is many times larger than it needs to be, and JPEG fixes it.
When it is the wrong target
Screenshots, logos, diagrams, anything with text or hard edges. JPEG's compression assumes smooth photographic gradients; on a sharp boundary it produces visible ringing, and the file may not even be smaller.
The test is simple: if the image has large areas of exactly one colour, JPEG is the wrong format.
The transparency question
JPEG cannot store an alpha channel. Every transparent pixel has to become something, and this tool makes it white — the common expectation, and the right default for a logo going onto a white page.
If your image sits on anything other than white, that decision is visible and wrong. Use WebP: it compresses like JPEG and keeps transparency.
Questions
What happens to transparent areas?+
They are filled with white, because JPEG has no alpha channel and something has to go behind them. If your image was designed to sit on a dark background, that white will be visible as a halo — convert to [WebP](/t/webp-converter) instead, which keeps transparency.
How much smaller does it get?+
For a photograph saved as PNG, dramatically — often 80–90%. For a screenshot, logo or diagram, sometimes not at all, and the result will look worse because JPEG smears sharp edges.
Will text stay readable?+
Less so. JPEG was designed for photographs and handles hard edges badly; small text picks up visible ringing. For anything with text, keep PNG or use lossless WebP.
Does this remove metadata?+
Yes, as a side effect. The output is built from decoded pixels, so any EXIF, colour profile or embedded comment is gone. Usually welcome for publishing — check with the [metadata viewer](/t/image-metadata-viewer) if you needed it kept.