Online mastering works on one finished stereo file. It adjusts tone, dynamics, and loudness for the whole thing at once, and it never reaches back into your mix. So preparing a mix for online mastering comes down to a single question: is this balance the balance you are willing to release?
Most prep checklists you will find answer that question for someone bouncing out of a DAW with a master chain they can switch off. If your file came from an AI remix, a cover, an extension, an add-vocals pass, or a quick bounce with no master chain at all, you have a different situation: one stereo file, no faders left, and no way to “leave headroom” after the fact. This is the ready-check for that file.
What Online Mastering Changes, and What It Leaves Alone
| Mastering can do this | Mastering cannot do this |
|---|---|
| Balance frequency response across the finished track | Rebalance individual stems or lift a buried vocal |
| Control dynamic range and smooth moderate level drift | Rewrite the arrangement, move sections, or fix timing |
| Push loudness and clarity toward a style or reference | Remove clipping and distortion already printed in the file |
| Make several tracks sit together as one release | Make a mix decision you have not made yet |
LumiMusic's AI Mastering page says the same thing in its own FAQ: mastering works on your finished stereo file, so problems baked into the mix stay in the master. That is the operating principle, not a disclaimer.
One assumption worth dropping before you start: louder is not the prize. Streaming platforms normalize playback level. Spotify adjusts tracks to -14 dB LUFS following the ITU 1770 standard and applies negative gain to masters that arrive louder than that. Apple's Digital Masters guidance describes Sound Check, the Apple Music feature that lets listeners hear all their songs at approximately the same loudness. Squeezing two more decibels out of a bounce does not make it louder for the listener. It costs you dynamics instead.
The Ready-Check Before You Upload
None of this needs a plugin, a meter, or a DAW. Play the bounce and answer six questions.
- Is this the balance you would release? Listen at a low, conversational volume. If you lose the lead vocal or the main hook at that level, the mix is not finished.
- Does anything crackle at the loudest moment? Find the biggest chorus, drop, or last hit. Fizz, crunch, or a grainy edge on peaks means clipping.
- Do sections jump in level? Move between intro, verse, and chorus. A small drift is normal. An obvious step up or down is an arrangement or mix issue.
- Does it already sound like a finished commercial release? Loud, dense, and consistently level throughout usually means limiting has already been applied.
- Is this the real final bounce? Not a screen recording, not a preview MP3 from another site, not a file you already pushed through a volume-boosting converter.
- Do you have the right to master this audio? Both the track and any reference file you plan to upload.
Then confirm the file itself will go through. Uploads accept MP3, WAV, FLAC, OGG, WebM, and M4A files up to 20 MB and eight minutes. If the track was generated on LumiMusic, you can pick it straight from your music library and skip the upload entirely, which also avoids an unnecessary re-encode.
All six clear means the file is ready. One failure means reading the next section before you spend anything.
Match the Symptom to the Decision
This is where AI bounces and simple DAW bounces behave differently from the multitrack sessions most prep guides assume. You cannot reopen a fader, so the only real choice is master now or go back and regenerate.
| What you hear | What it usually means | What to do |
|---|---|---|
| Crunch or fizz on the loudest hits | Clipping is printed into the file | Go back. Re-export or regenerate a clean bounce |
| The vocal disappears under the beat at low volume | Mix balance, not tone | Go back. Regenerate the vocal or rebuild it with Add Vocals |
| One section is clearly louder or quieter than the rest | Level jump between stitched or generated sections | Go back. Regenerate that section, then master the whole file |
| The join between the original and a generated continuation is obvious | The extension itself changed character | Fix the join first. Mastering evens tone, not identity |
| Already loud, dense, and flat from start to finish | The file is effectively pre-limited | Mastering is optional here, and will change little |
| Balanced but dull, thin, or quiet | Tonal, and this is what mastering is for | Send it |
| Muddy low end where kick and bass compete | Mix balance | Borderline. Mastering can shave, not separate |
| Harsh sibilance or a brittle top end on a cover | Vocal processing in the cover render | Send it, then judge the result against the original |
| The master sounds nothing like your reference | Style and reference are fighting | Switch to a style preset in the same genre |
The pattern is simple enough to hold in your head: tonal complaints are mastering's job, and balance complaints are not. “Dull” is a mastering problem. “I cannot hear the words” is not.
Clipping Is the One Failure Worth Going Back For
Clipping happens when a signal exceeds what the format can represent, so the tops of the waveform are cut off and the artifacts are printed into the file. Apple's Digital Masters document asks engineers to avoid clipping the signal outright, and explains that values pushed beyond 0 dBFS get cut off and create unpleasant, dissonant artifacts on playback. Nothing downstream undoes that. Turning the file down afterwards lowers the volume of the distortion, not the distortion.
What to do depends on where the bounce came from:
- An AI generation. Download the original result again instead of a copy you re-exported, converted, or “normalized” in another app. If the distortion is in the generation itself, regenerate that take.
- A DAW bounce. Pull the master fader down a few decibels and bypass the final limiter, then bounce again. A quiet input is not a problem for online mastering. A distorted one is.
- A layered bounce you assembled yourself. If a vocal over a beat clips only when both play, lower one of them before you bounce, not after.
There is no input loudness you need to hit. A clean bounce at a modest level gives mastering more to work with than a hot one that is already shaved flat.
When the File Already Sounds Mastered
Plenty of AI bounces arrive loud, dense, and even from beginning to end. That is not a defect, but it does change what a master can add. If the loudest chorus and the quietest verse feel like the same level, the loudness work is largely done.
Mastering that file is still reasonable when you need several tracks treated with the same style so they sit together on a release, when you want a tonal direction the bounce does not have, or when you need a WAV or FLAC export. Outside those cases, a second round of loudness processing mostly adds density.
If your track is thin rather than loud, that is a different job. A sparse voice memo sketch needs arrangement before it needs a master.
Choose a Style Preset or a Reference Track
LumiMusic offers eight mastering styles — Default, Pop, Hip-Hop, Electronic, Rock, R&B, Folk & Acoustic, and Metal — or a reference track you supply.
Use a style preset when your genre is obvious and you have no particular target sound in mind. It is the faster path, and it is the right default for most unmastered mixes. Pick Default when the track sits between genres or you are unsure; a mismatched genre preset is worse than a neutral one.
Use a reference track when you have one specific song whose tonal balance, loudness, or character you want to match closely. Choose it from the same genre and a comparable arrangement density. A sparse acoustic take referenced against a wall-of-sound electronic master will come back bright and crowded, because the reference is describing a song with far more going on in it. If your mastered result sounds unrelated to what you expected, the reference is usually the reason.
Set the output format — MP3, WAV, or FLAC — before you run the task, since it is part of the request rather than something you switch afterwards.
The left column is the send form: the bounce, Style or Reference, a preset, and the output format. The right column is where you play the result against the original at a matched level after the task finishes.
Each mastering task costs 10 credits, and credits are returned automatically if the task fails for a processing error. You can play and compare the mastered and original versions in the workspace either way; downloading the audio requires Starter, Pro, Unlimited, or active credit-pack access.
Worked Example: A Remix Bounce That Fails One Check
Treat this as a decision walkthrough rather than a recorded test. Imagine an authorized indie-pop remix bounce called Slow Ferry, 2:41, exported as a single MP3.
Running the ready-check:
- Balance: at low volume the chorus vocal vanishes under the synth pad.
- Clipping: none audible on the last chorus.
- Sections: consistent from intro to outro.
- Already mastered: no, it sounds quiet and slightly dull.
- File: the original render, 4 MB.
- Rights: the track is yours.
One failure, and it happens to be the failure mastering cannot touch. A master would make that quiet, dull file louder and brighter, and the vocal would still be buried — arguably more so, because a brighter pad competes harder. The correct move is to go back and regenerate with the vocal held forward, or rebuild the vocal separately, then run the ready-check again.
Now change one detail. If the vocal had been clearly audible and the only complaints were “quiet” and “dull,” every failing condition disappears and the file is ready as it stands. Same track, same length, opposite decision. That is the whole point of checking before you upload rather than after.
Judge the Master Before You Keep It
Compare the mastered version against the original at a matched level, not with the louder file simply sounding better because it is louder.
- Vocal: still sits where you left it, and is not pushed back by a brighter arrangement.
- Low end: firmer, not inflated into a boom that masks the kick.
- Loudest section: cleaner and bigger, with no new grit on peaks.
- Quietest section: still quieter than the chorus, not pumped up to match it.
- Character: recognizably the same record, with more polish.
If the master fails on vocal or low end, the mix was the problem after all. If it fails on character, try a different style or drop the reference.
Only Master Audio You Own or Are Authorized to Use
Master your own recordings, material you have permission to use, or files whose license allows the processing and release you have planned. The U.S. Copyright Office's guidance for musicians explains that a musical work and its sound recording are separate copyright-protected works, licensed separately. The same care applies to a reference track you upload: use it to guide tone, not as material you redistribute. Rules vary by jurisdiction and release plan, so unclear rights need qualified advice rather than a guess.
Frequently Asked Questions
Do I need to hit a loudness target before uploading?
No. There is no input number to chase. Aim for a clean, unclipped bounce you would be content to release if nothing else changed. Published figures such as Spotify's -14 dB LUFS integrated target and -1 dB True Peak ceiling describe the finished master, and playback normalization applies to it regardless.
Should I bypass the limiter on my master bus?
If the limiter exists only to make the bounce louder, bypass it and re-export. If it is genuinely part of the sound you want, leave it and expect mastering to change less. Either way, do not add a limiter in order to prepare a file for mastering.
Can mastering fix a vocal buried in the mix?
No. Mastering treats the finished stereo file as a whole and does not rebalance individual parts. Fix the balance, then master the result.
Conclusion
Preparing a mix for online mastering is a decision more than a checklist. Send the bounce when your complaints are tonal: dull, thin, quiet, or inconsistent with the rest of your release. Go back when your complaints are structural: a buried vocal, a clipped chorus, a section that jumps, or an arrangement that is not finished. Pick a style preset unless you have a genuine reference in the same genre, and judge the result at a matched level rather than by which file sounds louder.