Stem Prep Checklist for AI Mixing — Naming, Timing & Headroom | MegaMix AI
Start with clean exports and shared timing
Every stem should start at the same bar zero and use the same sample rate as the session, usually 44.1 kHz or 48 kHz at 24-bit. If the vocal stem starts 23 milliseconds late because you bounced from bar 2, the kick and vocal will feel smeared no matter how smart the mix engine is.
Leave 3 to 6 dB of peak headroom on each stem and avoid clipping. Inter-sample peaks that look “fine” on a DAW meter can turn into brittle 6 to 8 kHz fizz after later processing. Do not normalize every stem to 0 dBFS; that throws away the balance you already had.
If you recorded at 48 kHz, export at 48 kHz. Converting to 44.1 kHz just for upload adds a conversion you do not need yet.
Naming and grouping that actually helps
Use boring, consistent names: Lead Vocal, Adlibs, Hook Stack, Kick, Snare, Claps, Hats, 808, Bass, Pad, Guitar L, Guitar R. A folder of Track 12_finalFINAL2.wav files forces the mix to guess, and guesses show up as the adlib sitting louder than the lead.
Separate doubles and harmonies from the lead. If they live in one stereo file, a vocal-up request will also shove the stacks, and you will fight phase and level inside a single stem. For a trap hook, keep Lead, Doubles L/R, and Adlib as three files even if that feels extra.
Group only what truly shares a job. Kick and snare should not be one “Drums” stem unless you are mixing a two-track and that is all you have.
- Lead Vocal, Doubles, Harmonies, Adlibs as separate files.
- Kick, Snare/Clap, Hats, Percussion, rather than one drum bus unless necessary.
- 808 or bass separate from kick.
- FX and risers on their own stem so they can be quieter than the hook.
How much processing should be printed?
Print creative tone if it is part of the sound design: the Auto-Tune setting, the guitar amp, the vocal telephone filter in the bridge. Those are the record. Avoid printing heavy limiting, clipper-to-the-pins, or multiband compression on every stem unless that chain is truly the instrument.
A pop vocal with light compression and EQ printed is usually fine. A vocal with a maximizer hitting 6 dB of gain reduction is not, because harshness at 4 kHz becomes permanent. When in doubt, keep one cleaner alternate for vocals or bass so a “less harsh” revision can work without you rebuilding the session.
Print delay throws and reverb only if they are compositional. Leave mix-glue reverb off the stems if you want the AI pass to set depth.
If a stem already looks like a master, you have printed too much.
Final sanity checks before upload
Solo each stem for clicks, DC offset thumps, and accidental silence in the first bar. Remove long dead air at the start unless the pickup is intentional. A 12-second empty pre-roll on the vocal stem against drums that start immediately will make the song feel late.
Export a quick reference bounce of your rough mix at the same length. After the AI pass you want to compare direction, not rediscover the arrangement. Check that no stem contains a duplicate of another: two “lead vocals” 3 dB apart is a phase fight, not extra thickness.
Confirm file types are WAV, not a 192 kbps MP3 you emailed yourself from your phone. A lossy bounce trains the mix on codec smear you cannot EQ out later.
Stem export workflow from a typical DAW
Do this as a batch so you do not mix naming styles mid-export. Mute unused MIDI, freeze only if you must, and bounce from the same locator every time.
If you only have a two-track, you can still upload a rough mix to MegaMix AI. Stems simply give more independent control over vocal rides and FX.
- Set the export range from bar 1 (or the true start) through the last reverb tail, plus a short fade if needed.
- Select 24-bit WAV at the session sample rate with no extra limiter on the export bus.
- Bounce each group with identical start time; do not trim leading silence per track.
- Name files with a two-digit prefix if you like order: 01_Kick, 02_Snare, 03_LeadVox.
- Zip or drop the folder, then keep the rough mix bounce beside it for A/B.
Stem prep checklist before you hit upload
Run this list once. It is faster than diagnosing a muddy AI mix that was actually two bass stems fighting.
If any line fails, fix the export. Do not hope the mixer will magically ignore a clipped 808.
- All stems same start, same sample rate, same bit depth.
- Peaks generally below -3 dBFS; no red clips on individual files.
- No duplicate vocals or doubled instrumentals printed twice.
- Creative FX printed; mix-bus limiting not printed.
- Rough mix bounce saved for comparison.
- Lead, drums, and bass identifiable from filenames in under five seconds.
How to upload stems in MegaMix AI
Go to /app in your browser and upload the stems or the rough mix. MegaMix AI balances the session, then you refine with Josh. No DAW is required for that loop. Export WAV when the vocal sits and the low end is not a blob. New here? Start free trial or See plans.
After upload, your first Josh prompt should mention the files you care about by role: raise lead vocal about 1 dB in the chorus, keep adlibs 3 to 4 dB under the lead, shorten 808 sustain under the kick. Vague “fix my stems” wastes the naming work you just did.
If you still need a custom riser or downlifter, generate 2 to 12 seconds in MegaSFX AI at /megasfx and add it as its own stem next time instead of burying it inside a cluttered FX bus. Independent level control beats hoping a buried whoosh sits after the balance. Once stems are ready, the Mix AI hub covers the browser mix-and-master loop.
FAQ
Should I export mono or stereo stems?
Export mono for true mono sources like kick, snare, and a centered 808 when your DAW allows it. Keep stereo for pads, room mics, wide guitars, and vocal doubles that were recorded or processed as a pair.
How many stems is too many?
More stems help until two files contain the same performance. A 12 to 24 stem pop or trap session is usually plenty; Eighty micro-stems of the same pad at different plugin settings create overlap, not control.
Do I need separate instrumental and vocal stems?
Yes if you want independent vocal rides, dry/wet FX, and a clean instrumental for video. A single stereo mix-down is a valid MegaMix AI upload, but you cannot lift only the hook stack afterward.
What bit depth and sample rate should I use?
Match the session: commonly 44.1 kHz or 48 kHz at 24-bit. Do not upsample a 44.1 kHz project to 96 kHz for “quality” before AI mixing; you add filesize without adding musical information.
Should I include the mix bus compressor on every stem?
No. Mix-bus glue on every stem double-compresses when the AI sums them; Print instrument tone; leave the glue for the mix pass.
What if I only have a phone voice note plus a beat?
Export the beat as a stereo instrumental and the vocal as its own file, both from the same start time. Align them in any editor first if the voice note was recorded late, then upload those two files to /app. More on the Mix AI hub. See Start free trial or plans.