How loud should a short video be? The -14 LUFS rule, explained
Every platform turns your audio up or down to its own target before anyone hears it. Deliver at the wrong level and your clip comes out quiet next to the one before it, or crushed. Here are the numbers, what they mean, and the two mistakes that undo them.
Audio is the half of a short that nobody looks at and everybody feels. A clip that plays quieter than the one before it reads as amateur before a word lands; a clip that clips (the other kind of clipping, the distorted kind) gets scrolled past. Both are fixed by one number that the platforms have all quietly agreed on.
What LUFS is, in one paragraph
LUFS measures loudness the way ears hear it, averaged over the whole clip, rather than the height of the loudest spike. A peak meter tells you whether the signal distorts; a LUFS meter tells you how loud the thing feels. Platforms normalise on the second number: they measure your clip, then turn it up or down so it sits at their target. If you deliver quieter than the target they turn you up, and your noise floor comes up with you; if you deliver louder they turn you down, and any limiting you did to get there stays baked in.
| Platform | Loudness it normalises toward | True peak ceiling | What that means for you |
|---|---|---|---|
| YouTube and Shorts | about -14 LUFS | -1 dBTP | Louder than -14 is turned down; quieter is left quiet |
| TikTok | about -14 to -16 LUFS | -1 dBTP | Deliver at -14 and nothing is touched |
| Instagram Reels | about -14 LUFS | -1 dBTP | Same as above; music beds are the usual offender |
| Spotify (video podcasts) | -14 LUFS | -1 dBTP | Podcast-loud is already right |
| Apple Podcasts | -16 LUFS | -1 dBTP | A touch quieter; -14 is still turned down gracefully |
The two mistakes that undo a correct level
The first is the peak. A clip normalised to -14 LUFS can still have peaks above zero, and the compressed file the platform makes will overshoot them further. Keep the true peak at -1 dBTP or lower, and if the source was already brickwalled, leave more room, because the encoder adds up to a decibel on material that has no headroom left.
The second is the music bed. A bed mixed in after the normaliser pushes the whole clip up again; a bed mixed in before it steals loudness from the voice, so the platform turns the voice down to make room for the music. Mix the bed under the voice first, then normalise the result as one signal.
Checking a clip yourself
- Play your clip straight after someone else's in the same feed. If you reach for the volume, the level is wrong.
- Look at the waveform of the loudest second. A flat top means it was limited hard; the platform will keep that.
- Listen on a phone speaker, not headphones. Small speakers hide bass and expose the noise floor a quiet clip drags up.
Every rendered clip is measured and normalised to -14 LUFS with the true peak held under the ceiling, in two passes so the gain is one clean number rather than a pumping compressor. A source that already clips gets extra headroom automatically. Music beds are mixed under the voice before the measurement, never after.
Make a clip and play it after anything else in your feed. It will sit at the same level.
Want clips like this from your own footage? Try the AI video clipping tool free, no credit card.
Turn your long videos into viral shorts
SpikeReel finds your best moments, reframes them vertical, and burns in captions, automatically.
Start free