How to Score a Video Edit with Stems from Loudly
Most people treat background music as one track, drop the volume to 20 percent under a voiceover, and call it done. The result sounds like two unrelated things playing at once. Loudly’s stem export on paid plans lets you pull the mix apart, which fixes this properly.
Stems usually mean drums, bass, melodic elements, and vocals as separate files. When you control them individually, you can keep the rhythm audible and remove only the parts that clash with speech.
Why does a track fight a voiceover?
Because speech and music occupy the same frequency range. Conversation sits mostly between 300 Hz and 3 kHz, and so do pianos, guitars, and vocals.
Lowering the whole track reduces the good parts too. You end up with something that sounds quiet but still muddy. Removing the melodic mid-range leaves the drums and bass to carry energy while the voice owns the centre.
How do you decide which stems to keep?
Start with drums and drop everything else. If the edit still feels alive, you are done. Most talking-head videos need nothing more.
Then add back in order:
- Bass — adds weight, safe under voice, keep at full level.
- Percussion and texture — hats, shakers, vinyl noise. Adds movement without competing.
- Melodic elements — the risky layer. Add at minus 12 dB first and listen.
- Vocals in the music — almost always remove. Two voices at once is unintelligible.
How do you handle timing and structure?
A generated 60-second loop under a 4-minute video gets noticed by minute two. Two fixes help.
Cut the music at edit points. Change music when the scene changes, even if the new section is the same track at a different level. Listeners read the change as intentional.
Leave gaps. 10 to 15 seconds of voice-only at a key moment makes the music’s return feel like a decision instead of wallpaper.
What levels actually work?
Depends on the platform, but a starting point:
- Voice at minus 6 dB peak.
- Music full mix at minus 22 to minus 26 dB under speech.
- Drums-only version at minus 14 to minus 18 dB.
- Ducking of 3 to 4 dB with a 150 ms release if you use sidechain compression.
YouTube’s loudness target is around minus 14 LUFS integrated. Normalise at the end, not per-stem.
What are the common mistakes?
Normalising stems individually. That destroys the balance the generator produced. Normalise the sum.
Removing all low end. Music with no bass under speech sounds thin and amateurish on phone speakers.
Using more than four stems. If the exporter gives you eight layers, you do not need eight. More layers means more decisions and a worse mix.
Does the licence cover this?
Stem export and editing are covered on paid plans, and modified tracks still require the plan’s commercial licence. Free-tier downloads typically do not include stems, which is the practical reason most people upgrade.
If you are distributing on a platform with content ID, register the final mix, not the stems.
When should you not bother?
Single-shot videos under 60 seconds. The effort of a stem mix is not worth it, and a single track at low volume works fine.
Also skip it for any project where the music is the point — a music video or a reel where the track is foreground. Stems help music recede, not lead.
Quick Summary
- Speech and music compete between 300 Hz and 3 kHz, so remove mid-range melody rather than lowering everything.
- Keep drums first, add bass, add melody last at minus 12 dB, and remove music vocals entirely.
- Cut or change the music at edit points and leave 10 to 15 second gaps so it does not become wallpaper.
- Target music at minus 22 to minus 26 dB under voice and normalise the final mix, not individual stems.
- Stem export requires a paid Loudly plan and is not worth it for clips under 60 seconds.
