ace-studio-features

Files (1)

• SKILL.md

SKILL.md (first lines)

# ACE Studio features

# Tracks

### Track types

A track has no fixed type, and by default it is a blank track. It becomes what
you put on it — load a singing voice and it is a vocal track, load an instrument
and it is an instrument track. A project starts with a large pool of empty
tracks waiting to be filled.

| | Catalog | Driven by |
|---|---|---|
| Blank track | Blank | |
| Audio track | Audio | audio clip |
| Voice track | MIDI | MIDI + lyrics + vocal parameters |
| AI instrument track | MIDI | MIDI + articulations + instrument parameters |
| Plugin instrument track | MIDI | MIDI + MIDI CC |
| Master track | | |

- A MIDI track with no sound source is not silent — it falls back to a standard
  piano sample library. When the user says "it sounds like a piano", the answer
  is almost always that nothing is loaded, not that something broke.

- Every track except a blank track can host audio effects.

- **Unison Mode** belongs to voice tracks and AI instrument tracks only. It lets
  several voices or instruments load onto one track to produce a unison. Open
  the spread up when you want a wider stereo image.

### Audio effects

- ACE Studio ships basic effects, but prefer the third-party plugin effects the
  user has installed locally.

- Voice tracks and instrument tracks come with a basic effect chain already
  loaded. Add to it or strip it back as the work requires.

- **Room Effect** is a spatial effect that exists to serve Unison Mode: it
  models one acoustic space and places the unison speakers at different
  positions within it. Turn it on only when the track plays a background role.

- **To take an effect out of the chain, deactivate (disable) it — do not bypass
  it.** Bypass is for comparing a plugin in context, a listening call an agent
  does not make.

# Vocal Synth, AI Instruments, and instrument plugins

### Vocal Synth

Use Vocal Synth when the user wants to create a vocal. It is the sound source
that takes MIDI and lyrics, and synthesises a sung vocal through a chosen
speaker.

Recommended speakers:

| For | Speakers |
|---|---|
| Big pop vocals | Ember Rose, Rebecca, Bianca, Mangus |
| R&B, jazz | Jessica, Zalo |
| Lyrical, tender | Elirah, Emma |
| Epic, cinematic | Clara |

**Blend voice** creates a new voice by mixing several speakers. When the user
describes a voice they want, pick from the library first; reach for a blend only
if the library does not get there and they want something bespoke.

If the user wants their own voice, or a specific voice they supply, open the
voice library and point them at the **Cloning** tab to train it themselves. That
flow happens in a browser. Consult the wiki and walk them through it.

**Vocal Synth renders in the cloud, but there is a turbo mode** that moves most
of the model computation onto the local machine. Studio measures whether the
machine can support turbo mode the first time it launches: if the local compute
is enough it is on by default, and if the check fails it cannot be enabled.
Turbo mode turns on **pre-rendering** by default, so an edit triggers a render
immediately. With turbo mode and pre-rendering together, rendering is usually
finished before playback even starts.

- Without pre-rendering, rendering is triggered by playback.
- With pre-rendering, rendering is triggered by the edit.

### AI Instrument

When the user wants an acoustic instrument, prefer an AI Instrument — its
realism is far beyond a sample library.

AI Instruments do not render in real time, so each instrument type carries a
very small sample pack purely so MIDI input responds as you play. **That sample
is not what the model outputs.**

**String Section** is modelled on a string orchestra. Its speakers are Violins I,
Violins II, Violas, Celli and Basses. These support polyphony well, but keep the
written part idiomatic for the section it is on.

AI Instruments render in the cloud, with no turbo mode and no pre-rendering, so
rendering is always triggered by playback.

**Every other instrument is a solo instrument.** You can feed them polyphonic
material, but they do not behave the way a sample library does:

- Strings tolerate a little polyphony — but as one instrument bowing several
  strings, not as a section.
- Wind instruments cannot physically play polyphonically, so the model is forced
  to fake it. Do not do this.
- When a solo instrument needs several parts, split them across tracks instead.

Instruments named **Vintage** perform less well than the others.

**Articulations are marked on notes**, and each instrument type has its own set.
The default **smart mode** works out the articulation from the melody and applies
it in the synthesised result. Two articulations never appear under smart mode and
must be set explicitly:

- Pizzicato
- Mute

### Rendering — Vocal Synth and AI Instruments alike

From BeatMagic/acestudio_agent_plugin, MIT.