What Is the Difference Between Lossy and Lossless Compression?

Lossless guarantees reversibility. Compress and decompress, and you have the identical file — every pixel, exactly. This is the same guarantee a ZIP archive makes about a document, and the reason you can trust it for a master copy. The W3C specification for PNG puts the principle plainly: its filtering and compression "preserve all information".

Lossy guarantees nothing about the data and everything about the appearance. It discards information permanently, selected specifically because human vision is bad at noticing its absence, and the amount discarded is what a quality setting controls.

Note the asymmetry in the middle rows, because it is the whole practical story: lossless is excellent for the images lossy compression handles badly, and nearly useless for the images lossy compression handles brilliantly.

"Lossless is higher quality." No — for the same source, lossless and a high-quality lossy export look the same on a screen. Lossless is not a quality tier; it is a guarantee about the data. Saving a photograph as PNG does not improve it. It preserves it, at roughly ten times the size, with no visible benefit at normal viewing.

"Lossy means bad." An 80%-quality photograph is what almost every image you have ever seen on the web is, including the ones you thought looked excellent. The detail is genuinely gone; you cannot see that it is gone. Refusing lossy compression for web photography means shipping files ten times larger for a difference nobody perceives.

"I can convert lossy to lossless to fix it." The most expensive misconception, because it feels productive. Converting a JPEG to PNG gives you a lossless copy of a lossy image — the discarded detail is not in the file to recover. You get a much bigger file containing exactly the same visible picture, artifacts included.

"Lossless is always smaller than the original." Not necessarily. Lossless compression finds redundancy, and if there is none, the output can be marginally larger than the input plus its overhead. This is also why re-compressing an already-optimised image tends to grow it: the previous encoder's artifacts are themselves detail the new encoder must preserve. ToolForge's compressor watches for exactly this and keeps your original bytes, marking the file "Already optimised" rather than handing back something worse and bigger.

The decision follows from what the image is made of, not from how much you care about it.

Choose lossless when the image has large flat areas, hard edges or few distinct colours — screenshots, diagrams, logos, line art, charts, pixel art. Also whenever the file is a master you will edit again, or when it must survive multiple save cycles, or when an exact reproduction is a requirement rather than a preference.

Choose lossy for photographs and any continuous-tone image with fine detail throughout, which is where the savings are dramatic and the cost invisible. That covers essentially all photographic work destined for a screen — and the mechanism explains why: what gets discarded is chosen specifically to be hard for you to see.

The awkward middle case is a screenshot that contains a photograph, or a product shot on a flat white background. Try both; the file sizes will usually make the decision obvious, and if lossy wins on size, check the edges of the text before accepting it.

One more option that neither family covers: for logos and icons, the best answer is often not a raster format at all. SVG stores shapes rather than pixels, so it is sharp at any size and typically smaller than either.

Two modern developments complicate the neat split.

Formats do both. WebP and AVIF each have a lossy and a lossless mode, so "is this a lossy format?" is no longer answerable from the extension alone — it depends how the file was written. WebP lossless is the most useful example: it competes directly with PNG and Google measures it 26% smaller, while WebP lossy competes with JPEG at 25–34% smaller. One format, both jobs.

"Visually lossless" is a real and useful category. It means lossy compression tuned so the difference is imperceptible in normal viewing — data lost, appearance preserved. Most professional image delivery lives here. It is worth using the phrase precisely, because telling a client an image is "not compressed" when you exported at 85% is inaccurate, while "visually identical at viewing size" is both true and reassuring.

If you ever need to prove a round trip was genuinely lossless rather than merely convincing, compare checksums: hash the original and the decompressed result, and identical hashes settle it. Different hashes with an identical-looking image is the signature of visually lossless compression.

The same split exists elsewhere, and borrowing the intuition is useful.

Audio. FLAC is lossless, MP3 and AAC are lossy. Nobody thinks a FLAC file "sounds better than" a 320 kbps MP3 in casual listening — they think it is the archival copy. That is exactly the right relationship between PNG and a high-quality JPEG.

Archives. ZIP is lossless because it must be: a document with a few bytes discarded is corrupt, not slightly worse. Which is the useful test for your own images — ask whether an approximation is acceptable. For a photograph on a page, it plainly is. For a scanned contract, a barcode, a QR code or a medical image, it plainly is not, and lossy compression can make those unreadable rather than merely uglier.

The general rule: lossy compression is appropriate when the output is for human eyes, and inappropriate when something is going to be read — by a scanner, a decoder, or a court.