ElevenLabs Music stems: Creator 2 and 4, Pro 6, and the Sume route

ElevenLabs lists stems as a paid-tier feature of its music app: 2 and 4 stems on Creator, up to 6 on Pro. Sume's Music Router docs describe one mixed audio file and no stem output, so if you need to remix drums or vocals separately you need a vendor that exports stems. If you only need the music to sit under a voiceover, a Timeline render with duck_db does that without stems.

What ElevenLabs says

The ElevenLabs music page lists stems on Creator (2 and 4) and Pro (up to 6). The music page also mentions "stem editing" under its API section, but the API docs page we read does not mention stems, so confirm in the API reference that stems are exposed before you build on them. That distinction matters to a pipeline builder: a feature on a pricing grid is not a feature on an endpoint.

Three jobs people call "stems"

Most requests fall into one of three jobs, and only one of them truly needs separated tracks.

• Quieter music under speech: use the Timeline soundtrack.duck_db (0 to 20) and gain_db. No stems required.
• A version without drums for a calm cut: write it in the brief for a second take on Music Router, or take stems from a vendor that exports them.
• A karaoke or vocal-only file: that needs separation or a generator that outputs it. Sume has no such step.

The Sume path in practice

Generate with POST /v1/music-router/generate, then place the artifact as soundtrack.url in a Timeline 1.0 render. The soundtrack fields are url, gain_db, loop, fade_out_seconds (up to 10) and duck_db; duck_db needs a real audio spine, so it works when a voiceover is the spine and fails with duck_requires_audio_spine when the spine is silence.

To change a track's length first, use timeline audio: split slices up to 20 ranges, each a new durable file, for $0.01 per job.

Be honest about the gap

If stems are a hard requirement for your workflow, Sume's Music Router docs describe no way to meet it and this post will not talk around that. If the requirement is really "voice must stay clear", ducking is the shorter route. Verify output from either path by listening before it ships.

Pricing and tier names move; check the live pages before you buy a plan for stems alone.