Suno v5: getting sampleable stems, updated for v6.

Updated October 2026. Suno released the v6 family on September 9, 2026: v6, v6-wild and v6-mini (Suno). This guide was written for v5; the restraint method still applies to v6, and the stems part now covers Suno's own Get Stems tool.

Every new version of Suno is better at one thing: sounding finished. Cleaner mixes, fuller arrangements, more convincing vocals. If you are an artist chasing a complete track, that is progress. If you are a beatmaker who wants raw material to chop, a model that finishes the song for you is working against you.

The good news is that the fix has never been the model. It is the brief. The same discipline that pulled loops out of older versions still pulls them out of v5 and v6, you just have to push a little harder against a model that really wants to hand you a full release.

What changed and why it matters for sampling

Treat every model jump as a jump in confidence. Newer versions tend to arrange more, mix tighter, and commit harder to a full song. That polish is exactly what a beatmaker has to undo before anything is useful.

For sampling, "more finished" usually means less to work with. Busy arrangements leave no space for your drums. Loud, glued mixes are harder to chop cleanly. Confident vocals show up even when you did not ask for them. So the goal with v5 is not to chase its best output, it is to steer it toward its loosest, most exposed output.

Prompting v5 for restraint

Front-load the boundary. In my sessions the start of the prompt weighs more, so lead with the kind of source you want: fully instrumental, melodic and harmonic only, no chorus, no bridge, no full arrangement. Leave drums out of the prompt entirely, even as a no: naming them can pull them in. Then describe a source recording, not a song.

Ask for space in plain terms: sparse, exposed, one main instrument, room tone, long held sections. The more you talk about restraint, the less the model reaches for a finished production. Keep the mood single and specific so it does not overcompensate with drama.

Instrumental source recording, fully instrumental, melodic and harmonic only. Single Rhodes over soft upright bass, sparse and exposed, lots of space. Warm tape texture, mellow late-night mood, slow and unhurried, repeated 8-bar phrases, no chorus, no build.

Getting stems and sections you can isolate

Suno now has its own stem tool. Get Stems, in the menu of a song in your Workspace or Library, splits it into separate parts, vocals, drums, bass, keys and more, for Pro and Premier subscribers. Advanced Stem Separation goes further and costs credits: Auto Split (up to 12 stems, 50 credits per extraction), Split from Mix (one instrument plus the rest) and, on Premier only, Advanced Split (pick from nearly 100 instruments): Suno counts 10 credits per extraction, 20 in total for the two stems you get, the part you asked for and the rest (Suno). Credit costs change, so check them before a long session. In Suno Studio, the Multitrack export gives you every track as a stem (Get Stems, Studio export).

Keep three things apart. A sparse take is just a thin arrangement: everything is still in one mixed file, and a high-pass or a mute will not separate instruments that play at the same time. An extracted stem is the algorithm's best guess at one part, cleaner on a sparse take, with artifacts possible on dense passages. A Studio multitrack export gives you the tracks as Studio holds them. So keep prompting sparse, with one lead instrument over a simple bed: it gives the separation less to untangle, and it chops better even if you never split it.

Cleaning up before you chop

Newer output often arrives loud, wide, and drenched in tail. Before you touch a chopper, tame it. High-pass the rumble, pull the stereo toward mono to check the core, and trim reverb tails that will smear your chops.

Then do the part that has not changed in twenty years: audition for the strongest four or eight bars, not the whole track. One clean, repeatable phrase beats a beautiful section you can never loop.

When to prefer an older model

Newest is not automatically best for sampling. An older, rougher model can hand you exactly the looseness you want: thinner mixes, more accidents, more exposed single instruments. Those "flaws" are chop fuel.

Run the same brief through more than one version and keep the take that fights you least. Sometimes that is v6 or v5 on a restrained prompt, sometimes v6-wild, which Suno describes as less predictable and more varied, sometimes an older model doing less, if your plan still offers it. Judge the material, not the version number.

Let Crate Machine handle the prompt

Writing a restraint-first brief every time gets tedious. Crate Machine builds the structured prompt for you, phrased positively (it describes the source it wants rather than listing exclusions), so you spend your time chopping instead of wording. Feed it a genre, era, mood, and one lead instrument, and it hands back a prompt aimed at source material rather than a finished song.

It is the same logic across every Suno version, so it keeps working when the model changes under you. Point it at v5 or v6, keep the parts you can use, and move on.

FAQ

---

For the prompt anatomy that makes any of this work, read how to use Suno for sampling: the prompt structure.