Generation loss is the cumulative degradation from repeated lossy encoding. Each pass discards information, and the next pass starts from what the previous one left — including its artifacts, which it then spends bits faithfully preserving.
The effect is not linear and it is not subtle after a few rounds. The classic demonstration is a video re-uploaded to a platform dozens of times: after enough generations the picture is mush and the audio a warble, from nothing but repeated compression of an originally clean source.
In ordinary use it arrives in less dramatic forms:
- A clip is exported from an editor, compressed to fit a chat app, forwarded, saved by the recipient, and re-sent. Four generations.
- A screen recording is compressed, uploaded to a platform that transcodes it, then downloaded and compressed again for email. Three.
- A file is compressed twice because the first attempt was still slightly too large. Two, where one would have done.
That last case is the avoidable one and the most common. Compressing a compressed file to shave off a further 10% costs a full generation for very little size. Going back to the original and compressing once to the right target costs nothing extra.
The defences are simple: keep the original, decide the final target before encoding, and prefer remuxing to transcoding whenever the goal is compatibility rather than size.