# AI tasks

> What AI can do for your music in Maru, where to try each task, and where to go when you want to build with it.

Canonical: https://maru.audio/help/ai-tasks
Documentation index: https://maru.audio/help/llms.txt

Some musical jobs are hard to do by hand and surprisingly good with a trained
model: turning a hummed melody into notes, pulling the vocal out of a song,
finding the downbeat, or using your hands as a controller. Maru treats each of
these as a **task**. You choose what you want done; the module you are using
decides which model does it and where it runs.

## Choose a task

| You want to | Task | Try it | Read |
| --- | --- | --- | --- |
| Make a new sound from a description | Generate audio | Instruments with a **Generate** action | [Generative features](https://maru.audio/help/generative-features.md) |
| Turn singing, playing or a recording into MIDI | Audio to MIDI | [Pitchy](https://maru.audio/modules/pitchy) | [Turn audio into MIDI](https://maru.audio/help/turn-audio-into-midi.md) |
| Pull vocals, drums or bass out of a song | Separate stems | [StemSep](https://maru.audio/modules/stemsep) | [Split a song into stems](https://maru.audio/help/split-a-song-into-stems.md) |
| Find the beats and downbeats in a track | Beats and tempo | [StemSep](https://maru.audio/modules/stemsep), **Beat grid** | [Follow the beat](https://maru.audio/help/follow-the-beat.md) |
| Control sound with your hands, face or body | Camera control | [HandMod](https://maru.audio/modules/hand-mod), [Hand Chord](https://maru.audio/modules/hand-chord) | [Play with your camera](https://maru.audio/help/play-with-your-camera.md) |

## Where to do what

- **Try a task** on a module's page. Nothing to install; some tasks download a
  model the first time and ask you first.
- **Understand it** in [Learn](https://maru.audio/learn), which explains how each technique works
  and where it breaks.
- **Build with it** in [Studio](https://maru.audio/studio): remix a module that already does the
  task, or ask your coding agent to add the task to your own module. Each
  article below includes a prompt you can give your agent.
- **Write the code** with the developer guides in
  [AI tasks in your module](https://maru.audio/docs/ai-tasks.md).
- **Compare models** in the [model directory](https://maru.audio/models), and read why Maru works
  on these tasks on [AI for music](https://maru.audio/ai).

## On your device or on a server

Each task runs either on your device or on a server, and the module tells you
which. On-device tasks keep your audio and camera picture on your computer, but
may download a model first and need a reasonably recent browser. Server tasks
upload the audio, can use a larger model and may use credits. Check the method a
module shows before you process an unreleased recording.

## Where tasks work today

Tasks run on module pages on the website. Desktop plug-in builds (VST3, AU,
CLAP) play what you made with a task, such as a generated sound or a separated
stem, but cannot run most tasks themselves yet. Live pitch tracking is the
exception: it runs inside the module, so it works in every build.