music-generation

Music Generation (optional / legacy)

Status: optional, legacy path. The default podcast-creator pipeline does not generate music. The standard route mixes a pre-made brand music kit into the episode, and that is handled entirely by the audio-mixing skill (resolved from the profile's music_kit reference or the music_kit_path machine config). Use this skill only when a brand kit is unavailable and a quick generated bed is explicitly wanted. Generated music is non-deterministic, Lyria is experimental, and the result will not carry the show's branded sound. Prefer the kit.

Generate a ~30-second ambient background clip using the Lyria model via the generate_content API.

Interpreter (R70). generate_music.py imports google-genai, which lives in the orchestrator's uv-managed venv — run it through the venv interpreter the orchestrator resolved in Step 0 ("$PODCAST_PY" <script>), not bare python3. The python3 … in the example is shorthand; substitute the resolved venv interpreter. See podcast-studio SKILL.md Step 0.

Invocation

python3 "${CLAUDE_SKILL_DIR}/scripts/generate_music.py" --workspace <run-dir> --mood tech

Arguments

Argument	Default	Description
--workspace	workspace	Run directory; the clip is written under its audio/music/ subdir.
--mood	default	Music mood (see moods below). default maps to tech.

Moods

Mood	Style
tech (default)	Clean synths, electronic pulse, forward-looking innovative sound
chill	Soft pads, gentle electric piano, lo-fi warmth, non-distracting
debate	Building tension, confident synth brass, panel-discussion opener

What it does

1. Selects the music prompt based on --mood.
2. Calls the generate_content API with the lyria-3-pro-preview model.
3. Generates a ~30-second music clip.
4. Saves it as MP3 to {workspace}/audio/music/background.mp3.
5. Handles failures gracefully — the pipeline continues without generated music.

Dependencies

• google-genai (>= 2.0.0)

Output

• File: {workspace}/audio/music/background.mp3
• Format: MP3
• Duration: ~30 seconds

Fallback

Lyria is experimental. If generation fails with a policy error or returns no audio, the script retries once with a simpler fallback prompt: "Create a 30-second simple ambient background track. Instrumental only, calm and neutral."

If the fallback attempt also fails, the pipeline proceeds without background music — the audio-mixing step handles a missing music bed gracefully.