Smaller, and every bit comes back.
Lossless compression only removes repetition and predictability, so unpacking gives back the exact original. Here your browser packs a file, unpacks it, and proves it with a SHA-256 fingerprint of both ends.
PNG doesn't compress pixels directly. Each row is first rewritten as the difference from a guess: the pixel to the left, the one above, their average, or Paeth's pick of the three. Good guesses leave mostly zeros, and zeros compress well. These are the five real PNG filters run on a test image, then packed with the same DEFLATE that PNG uses.
Running all five filters…
A general compressor looks for repeated byte strings, and recorded sound almost never repeats exactly. FLAC instead predicts each sample from the last few, using the fixed polynomial predictors in its specification, and stores the error with a Rice code. This runs both on one second of generated 16-bit sound.
The best predictor here, order 3, needs 3.25 bits per sample, 4.9× smaller than WAV, while gzip only reaches 15.25. Real FLAC encoders also try longer custom predictors and stereo tricks, and keep whichever is smallest for each block. Rice coding here uses the best parameter per 4,096-sample block.
Lossless: no data lost, perfect reconstruction, larger files. Lossy: data permanently discarded, smaller files, slight quality reduction.
Yes, PNG is always lossless. Every pixel is perfectly preserved when saving as PNG, unlike JPEG.
Yes, ZIP is always lossless. Files inside a ZIP archive are exactly identical to the originals when extracted.
For archiving or editing: use lossless (PNG or TIFF). For web sharing: JPEG or WebP lossy is fine as they look almost identical at good settings.
FLAC is the most popular lossless audio format. WAV and AIFF are also lossless but larger. Apple ALAC is Apple's lossless format.