A generated song already arrives mixed, which is fine for quick use but limiting the moment you want to change the balance, add your own instrument, or fix one element without touching the rest. Stems solve this by giving you the individual pieces the mix was built from.
Step 1: separate and export the stems
In Studio, run stem separation on your finished song to split it into six parts: vocals, drums, bass, guitar, piano and other. Export as FLAC rather than a compressed format if you plan to do real mixing work — lossless audio holds up better under EQ and compression than a lossy export, especially if you're layering it against other lossless material.
Step 2: line everything up in the DAW
Import all six stems into a new session as separate tracks, aligned to the same start point (they will already be in sync since they came from the same source render — the important part is not accidentally offsetting one during import). Confirm the session tempo matches the track if you plan to add MIDI instruments or loops that need to lock to the grid.
Step 3: mix like a multitrack session
From here it is a normal mixing workflow, just starting from AI-generated source material instead of individually recorded instruments.
- EQ each stem to carve space — cut competing frequencies between vocals and other melodic stems rather than boosting everything.
- Rebalance levels to taste; the AI's original mix is a starting point, not a fixed target.
- Mute or replace any stem you don't want — swap the AI drums for your own sample library, for instance, using the 'other' or bass stems as a reference for arrangement.
- Add your own effects (reverb, saturation, sidechain compression) exactly as you would on any other multitrack recording.
When to bring in MIDI instead
If you want to replace an instrument entirely rather than just remix its recorded stem — say, playing the piano part on your own virtual instrument — export that part as MIDI instead of audio, so you get the notes rather than a fixed recording to reprocess.