How to Prepare Your Stems for AI Mixing

The time you put into preparing your stems for mixing goes a long way when it comes to the business end of the mixing process.

This guide breaks down how to spend around five minutes preparing your stems to get the most out of them within Automix.

The principles here apply regardless of which DAW you use. If you want the step-by-step export process for your specific tool, then please check out how this can be done within Ableton Live, FL Studio, Apple Logic Pro and GarageBand. 

The following post is about the overarching principles that you should adhere to, whichever DAW you are using.

What Automix is working with

When you upload stems to Automix, the AI analyses each file individually and in relation to everything else in the session. It detects where elements are masking each other. It makes EQ, compression, panning and reverb decisions based on what it hears and what you tell it about each track.

The quality of those decisions depends entirely on the quality of what is given. Clean, clearly defined stems with accurate categorisation give it the best possible starting point. Cluttered, poorly formatted, or miscategorised stems produce results that reflect those problems.

This is not a limitation of the technology - it is how any mixing process works. Bad quality in, bad quality out applies to AI mixing just as much as it does to a human mixing engineer.

Separate stems beat grouped stems where possible

When you export a full drum group as a single stereo stem - kick, snare, overheads and room mics all bounced together - Automix processes it as one element. Every EQ and compression decision gets applied to the whole group. You lose precision on every instrument inside it.

When you export each drum element as its own stem, Automix can treat them independently. It manages the kick's relationship with the bass, controls the snare's dynamics on its own terms, and adjusts the overheads without affecting the close mics. The more defined each element is, the more targeted the processing can be.

The same logic applies across the whole session. A single guitar stem containing rhythm and lead parts gives Automix less to work with than two separate stems. A vocal stem with lead and harmonies bounced together is harder to balance than stems uploaded individually.

If your session is built with separate stems for each instrument or part, export them that way. If you are working with AI-generated audio or pre-made grouped stems, use what you have - Automix handles grouped stems well. But if the choice is yours, separate stems are worth the extra export time.

Format and technical settings

Export as WAV. Not MP3, not AAC, not anything compressed. Compressed formats would introduce artefacts that affect how the AI reads the audio. Automix only accepts WAV files for this reason.

Use 16-bit, 44.1 kHz as your standard settings, as this matches the output specs. There is no meaningful benefit to uploading 32-bit float stems; this just increases file sizes.

Keep levels sensible. Your stems do not need to be at any specific level before upload - Automix handles gain staging as part of the mixing process. But if a track is clipping, address that before you export. If a track is so quiet that it is barely registering, fix that too. The AI is reading the audio as a signal - give it a clean, stable signal to read.

Start everything at 0:00

Every stem needs to start at the same point in the timeline. This sounds obvious, but it is the most common issue in sessions that come back with timing problems.

If your kick starts at bar 1 and your vocal starts at bar 5 because there is nothing happening before then, export the vocal with four bars of silence at the start. When all stems begin at 0:00 and run to the same endpoint, Automix knows exactly how every element relates to every other element in time. A vocal that starts four seconds in, with four seconds of audio missing from the front, will come back with the timing relationship between the vocal and the rest of the mix broken.

Export everything from the very start of the project to the very end.

Effects - what to keep, what to leave off

Creative effects that are part of the sound - a specific reverb on a guitar that defines its character, a delay on a vocal that is integral to the arrangement, a saturator on a bass that is part of the tone - leave them on. Export with those effects remaining. If the reverb is part of the instrument, it is part of the stem.

Corrective processing that you have applied as a starting point - a high-pass filter on everything, a rough compressor to tame a particularly dynamic performance, some EQ to fix a problem - you can leave these on or take them off. Automix will apply its own corrective processing. If you have already done something useful, leave it. If you are not sure whether it is helping or hurting, take it off and let the AI start from the raw audio.

One thing to avoid: heavy limiting or mastering-style processing on individual stems before upload. If stems are already heavily limited, Automix has less dynamic range to work with at the mixing stage, and the result will reflect that.

Naming your tracks

Name every stem before you export. Not Track 1, Track 2, Track 3 - actual descriptive names that tell you what the track is.

This matters for two reasons. First, you will be assigning an instrument category to each stem in Automix, and doing that accurately is one of the most important things you can do to improve your result. If you are looking at a list of files called Track 1 through Track 12, you are slowing yourself down and increasing the chance of misassigning something. If the files are called Kick, Snare, Overheads, Lead Vocal, Acoustic Guitar and so on, the process is fast and accurate.

Second, descriptive names make the project file download significantly more useful. When you open the Ableton Live Set or DAWproject file in your DAW, you want to know what every channel is immediately. Well-named stems make that effortless.

Categorising stems correctly in Automix

Once your stems are uploaded, you assign an instrument category to each one. This is the step that has the biggest impact on your result after stem quality itself, and it is the step most commonly rushed.

Automix uses the category to understand what role each element plays in the mix. The processing applied to a kick drum is different to the processing applied to a synth pad. Getting this wrong means Automix is making decisions based on incorrect information.

A few of the less obvious ones worth knowing:

Toms go under Percussion, not Drums. Toms play a different role to the core kit elements - they are more transient, less constant - and Percussion is the right category for them.

Hi-hats go under Cymbals, not Drums.

Claps go under Snares.

The Drums category is for a full mixed kit stem only. If you have exported individual drum elements, use the specific categories for each one.

Bells and synth bells go under Cymbals.

Vinyl crackle, noise and atmospheric FX go under FX or Other. Do not force these into an instrument category.

Lead vocal and backing vocals should be separate stems where possible. It gives Automix targeted control over the most important element in most mixes. For more on getting vocal stems right, How to Mix Vocals Using AI covers the vocal-specific decisions in more detail.

Automix features an autodetect functionality that automatically selects instruments for you as suggestions. These can be switched if they do not match your intended choices.

For the full breakdown of drum stem categorisation, How to Mix Drums Using AI goes into more depth on every drum element.

It’s worth taking your time to ensure this step is actioned correctly. It is the highest-value thing you can do once your stems are uploaded.

Setting Importance before you generate

Before you generate your preview, set the Importance level for each stem - low, medium or high. This tells Automix how prominent each element should be in the finished mix. It shapes the processing chain across the board, not just the volume.

Think about what leads in your track and what supports it. Lead vocal, main riff, the part a listener's ear follows - high. Rhythm guitars, pads, harmonic layers - medium. Atmospheres, subtle textures, background elements - low.

If your first preview does not sound right, Importance settings are the first thing to revisit. Adjust and regenerate - the preview will not update until you do.

For a full breakdown of how to use Importance settings and everything else in Automix to get the best result, How to Get the Most Out of Automix is the place to go after this.

The short version

Export individual tracks where you can. WAV, 16-bit. Everything starts at 0:00. Keep creative effects, remove heavy limiting. Name every stem clearly. Categorise accurately - especially toms as Percussion and hi-hats as Cymbals. Set Importance before you generate.

Five minutes of preparation before you upload. A noticeably better mix when it comes back.

Head over to Automix today to try out AI mixing. You can preview mixes of up to 16 stems for free, with 32 stem mixing available as part of Automix Pro.