> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gridshift.studio/llms.txt
> Use this file to discover all available pages before exploring further.
# Stems
> Splits audio into stems, such as vocals, drums and bass, with AI models that run on your Mac: onto tracks of their own, or as layers inside one sample. Select the clip, press ⌘K and run Split Clip into Stems….
## Split a clip into stem tracks
Select an audio clip, press `⌘K` and run **Split Clip into Stems…** (search for "separate" or "stems"). Choose **Song Stems** or **Drum Elements**.
## Rebalance stems inside one sample
Open the clip in the [Sample Editor](/samples/sample-editor.md) (`⌥X`), press `⌘3` for **Stems** mode and press **Generate Stems**. Then drag each lane's volume slider, or press its **S** to solo it.
## Good to know
* **The split result is a group track** with one child track per stem, time-aligned with the original clip. The original clip stays, muted: unmute it to compare or to go back.
* **Each stem is a regular sample track.** Apply effects, time-stretch, edit it in Focus Mode, or use it as a sidechain source like any other track. See [Groups](/tracks/groups.md).
* **Stems mode keeps everything on the original track.** It bakes the same models against the sample, so you mix the stems as one instrument instead of arranging them separately.
* **Audio only.** With a MIDI clip selected, the Command Palette shows **Audio Clip Required** instead.
* **Models download on first use and run locally**; no audio is uploaded.
> **Details**
>
> * **In Stems mode the waveform gives way to a stack of stem lanes.** A lane slider at zero mutes it; double-click resets it to 100 %.
> * **The Algorithm picker selects the model**, **Source Stems** or **Drum Elements**, and is disabled while a bake runs. While a bake runs, the body shows its progress; once it finishes, the lanes appear.
> * **A sample stores one set of stems at a time**, so switching to a kind that hasn't been baked drops back to **Generate Stems**.
> * **The slice letter keys keep working in Stems mode.** The navigation gestures of the other modes do not.
> * **Palette commands for Stems mode**: **Generate Stems in Sample**, **Solo Stem in Sample** and **Set Stem Level:** followed by the stem's name (in %, 100 = as separated, 0 = silent). **Split Clip into Stems…** and **Download Separation Model** sit in the Audio Tools group.
> * **The models are third-party.** Gridshift integrates them for source separation, and the specific algorithms may change as better models become available.
> **Models**
>
> | Split option (Stems mode name) | Model | Output stems |
> | - | - | - |
> | **Song Stems** (**Source Stems**) | HTDemucs | Vocals, Drums, Bass, Other |
> | **Drum Elements** | DrumSep | Kick, Snare, Cymbals |
> **Which flow to use**
>
> | Use this flow | When |
> | - | - |
> | **Split Clip into Stems…** | You want each stem on its own track for independent arrangement, effects and routing |
> | **Stems** mode in the Sample Editor | You want to rebalance the parts inside one sample (e.g. lower the vocals on a loop) without adding tracks |
> **Stems mode states**
>
> | State | What you see |
> | - | - |
> | Model not installed | **AI Model Required** and a **Download Model** button that opens the model download |
> | Model installed, no stems yet | **Stems Not Generated** and a **Generate Stems** button, using the model chosen in the **Algorithm** picker |
> | Stems present | The lane stack |
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.