Lossy vs lossless: what a format conversion really costs you

Ask a converter to turn one file into another and it will, almost always, say yes. The question it won't answer is what got thrown away in the process. Every format encodes a set of assumptions about what matters — and a conversion is the moment two sets of assumptions collide. This is a field guide to the collisions that actually cost people work.

Lossy and lossless are properties of a path, not a format

The usual framing — "PNG is lossless, JPEG is lossy" — is true but not useful, because loss happens on the path, not in the destination file.

• Lossless → lossless (PNG → WebP-lossless, WAV → FLAC): safe, fully reversible. The only thing you're trading is file size and compatibility.
• Lossless → lossy (PNG → JPEG, WAV → MP3): you pay a quality cost exactly once, and you choose how big it is. Done at a sensible setting, usually invisible. Irreversible.
• Lossy → lossless (JPEG → PNG): the trap. The PNG is "lossless" but it faithfully preserves the JPEG's compression artefacts — at several times the file size. You get the downsides of both formats. People do this constantly thinking they're upgrading the image.
• Lossy → lossy (JPEG → WebP → JPEG…): every hop recompresses and adds a new layer of damage. This is generation loss.

Generation loss compounds quietly

One JPEG save at quality 85 is fine. The same image saved-as-JPEG five times — cropped here, resized there, run through a messaging app that recompresses uploads — is visibly worse: ringing around text, blotchy flat areas, smeared edges. Each pass looks like a no-op; the sum isn't.

The rule that avoids it: keep one master, export many times. Edit the original (or a lossless copy of it) and generate a fresh lossy export for each destination. Never treat an export as the new source.

The tradeoffs nobody mentions

Transparency

JPEG has no alpha channel. Convert a PNG logo with a transparent background to JPEG and the transparency is composited onto a background colour — usually white, sometimes black, occasionally whatever the library felt like. There is no warning; there is just a logo in a white box on your dark-themed site. If you need transparency and smaller files, the move is PNG → WebP or AVIF, not PNG → JPEG.

Colour depth and colour space

Converting 10-bit or wide-gamut sources (HEIC from a phone, ProRes video, RAW exports) into 8-bit sRGB formats clips what the file can represent. Smooth sky gradients pick up banding; saturated colours flatten. Sometimes that's an acceptable price for compatibility — but it should be a decision, not a surprise. A good pipeline converts colour properly (with an actual profile transform) instead of copying raw pixel values.

Resolution isn't quality

Upscaling a 800×600 image to 4K in a converter adds pixels, not detail. The same goes for re-encoding a 128 kbps MP3 as a 320 kbps MP3 — the new file is 2.5× larger and sounds identical, because the information was already gone. If a conversion increases fidelity numbers without a better source, it's only increasing the file size.

Documents have their own version of all this

PDF → Word is a reconstruction, not a conversion: layout is inferred, and complex tables or multi-column pages come back scrambled. Flattening a layered file (PSD → PNG, SVG → PNG) bakes in one rendering and discards the editability forever. Rasterising an SVG also freezes it at one resolution — the whole point of the format gone in one export.

A short decision checklist

1. What does the destination actually require? Convert for a concrete consumer, not "just in case".
2. Is the path lossy? If yes, pick the quality setting deliberately and do the lossy step exactly once, last.
3. Does the source have anything the target can't hold — transparency, layers, wide gamut, vector data, metadata? Decide what happens to each before converting, not after.
4. Keep the original. Storage is cheap; a re-shoot isn't.

Most of this is just discipline, but the tooling matters too: defaults chosen by someone who has seen the failure modes beat defaults chosen to minimise server load. That's the standard we hold MorphTo to — conversions that surface these tradeoffs instead of burying them, whether you're converting one file in the browser or wiring the API into a pipeline.