Skip to content

usage.md

Latest commit

 

History

History
84 lines (62 loc) · 3.74 KB

usage.md

File metadata and controls

84 lines (62 loc) · 3.74 KB
Raw

Usage Guide

Generate a Loop

  1. Open the Text to MIDI tab.
  2. Choose the musical Key and Scale.
  3. Describe the musical idea, instrumentation, rhythm, or mood.
  4. Select a Provider and Model.
  5. Adjust the model controls that are visible.
  6. Select a SoundFont if audio playback is configured.
  7. Click Generate Loop.

The completed view provides:

  • Download Generated MIDI: Import the loop into a DAW.
  • Playback: Play rendered audio when audio synthesis succeeds.
  • MIDI Visualization: An interactive four-bar piano roll powered by Plotly.
  • Status: Concise feedback on provider, parsing, rendering, or configuration issues.

A useful description is specific without trying to reproduce the entire system prompt. For example:

warm neo-soul electric piano chords with syncopated upper extensions and a simple bass movement

The selected key and scale are added to the request automatically.

What happens during generation

flowchart LR
    Input[Loop description and settings] --> UI[Conductor Main]
    UI --> Core[Conductor Core generation engine]
    Core --> Provider[Selected model provider]
    Provider --> Core
    Core --> Save[Save MIDI and generation metadata]
    Save --> Visualize[Build piano-roll visualization]
    Save --> Audio{Audio tools available?}
    Audio -->|Yes| Render[Render playback audio]
    Audio -->|No| MidiOnly[MIDI remains available]
    Visualize --> Results[Completed view]
    Render --> Results
    MidiOnly --> Results
Loading

Conductor Main collects the settings and presents the results, while Core owns provider communication, structured response parsing, MIDI creation, persistence, and optional audio rendering.

Model-Specific Controls

Conductor Main reads packaged model metadata and adapts its controls when the provider or model changes:

Control Behavior
Temperature Shown for models that accept sampling temperature
Reasoning Toggle used by supported legacy Anthropic and Google models
Reasoning Effort Model-specific choices such as minimal, low, high, or xhigh

Changing providers resets the model to a valid choice and refreshes dependent controls. A hidden control is intentionally unavailable for that model rather than missing from the installation.

Model labels show input and output prices per one million tokens when pricing metadata is available. The saved generation records the provider-reported cost; local Ollama generations normally have zero API cost.

Prompt Editor

The Prompt Editor tab displays the current loop-generation system prompt. Saving creates or updates the app-owned override at ~/.conductor/main/Prompts/loop gen.txt. Subsequent generations use that override instead of Core's packaged default.

The prompt defines the structured loop contract, timing conventions, and broad musical guidance. Make targeted changes and keep the required output schema intact; an invalid or ambiguous schema can cause provider parsing failures.

Deleting the override file returns the app to Core's packaged prompt.

Stop Waiting Behavior

Generation runs in a background thread so Gradio can continue yielding status updates. Stop Waiting detaches the UI from the current wait, but it cannot cancel a provider request already in flight. The provider may still finish and incur cost after the UI stops displaying progress.

flowchart TD
    Start[Generate Loop] --> Request[Provider request begins]
    Request --> Waiting{Still waiting?}
    Waiting -->|Provider finishes| Results[Show generated results]
    Waiting -->|Stop Waiting clicked| Detached[UI stops showing progress]
    Detached --> InFlight[Provider request may remain in flight]
    InFlight --> Cost[Request may finish and incur cost]
Loading