dreamina-canvas-generate-audio

Dreamina Canvas Audio Node

This Skill owns the Canvas audio node. The two modes (tts, music) have opposite field requirements, and a missing or wrongly-applied field is rejected with cli.audio_music_model_required (exit code 2) for music or with the mode-specific contract for TTS.

Reference detail is in references/audio-node-contract.md.

When to use

• The caller wants to synthesize speech from text (TTS).
• The caller wants to generate a music clip from a prompt (music).
• The caller wants to switch an existing audio node from one mode to the other and needs the field set explained.

When not to use

• For approving or running generation — that is dreamina-canvas-quote-and-run.
• For the visual track of an audio-bearing composition — that is dreamina-canvas-manage-timeline.

TTS

dreamina-canvas --format json node create audio \
  --prompt "<text to speak>" \
  --mode tts \
  --voice-name "<discovered-voice>"

• --voice-name is mandatory. The CLI resolves the public name into the authoritative server-side identifier.
• --model is forbidden for TTS; the model is selected from the voice's underlying TTS model, not user-supplied.

Music

dreamina-canvas --format json node create audio \
  --prompt "<music prompt>" \
  --mode music \
  --model "<music-model-from-discovery>" \
  --duration 30

• --model is mandatory. Take it from the live full audio discovery shape selected by schema (model search --detail full or model list), filtered to a music mode entry.
• --voice-name is forbidden for music.
• --duration must be > 0 seconds; default 30.
• The CLI refuses to default the music model. A missing --model returns cli.audio_music_model_required (exit code 2). Do not retry with a name remembered from a different environment.

Field exclusivity

Mode	Required	Forbidden
tts	--voice-name	--model
music	--model, --duration > 0	--voice-name

--count is not accepted on audio nodes — a single output only.

Generation is a full replace

Same as image / video: touching generation flags means submitting the complete new block. Metadata is sparse. --clear-generation is the opt-out and is mutually exclusive with generation flags.

# Sparse metadata edit; generation remains unchanged
dreamina-canvas --format json node edit audio \
  --node-id <nodeId> --title "<new title>"

Paid execution is not this Skill's job

Hand off to dreamina-canvas-quote-and-run. This Skill never calls --run, never sets --credit-ceiling or --credit-token, never persists tokens.

Failure → recovery

Failure	requiredAction	What to do next
cli.audio_music_model_required (exit 2)	none	Re-run the schema-selected full audio discovery command, pick a music model, retry.
--voice-name rejected on TTS	none	Re-run voice list; add --language only if schema declares it.
--model passed on TTS (exit 2)	none	Remove --model; let the voice drive the model.
--voice-name passed on music (exit 2)	none	Remove --voice-name; supply --model instead.
--count passed on audio (exit 2)	none	Remove --count; audio is single-output.
Missing --duration on music (exit 2)	none	Pass --duration from the discovery bound.

What this Skill will not do

• Approve spend.
• Persist tokens, signed URLs, cookies, or session material.
• Accept --count on audio nodes.
• Use a remembered voice or music-model name across environments.
• Implicit-default the music model.

Policy

Invocation requires:

• Confirmed installation of dreamina-canvas
• Live voice list (for TTS) or live audio model discovery (for music)
• For paid execution: explicit handoff to dreamina-canvas-quote-and-run

Forbids:

• Calling --run from this Skill
• Passing --count on audio nodes
• Supplying --model on TTS or --voice-name on music
• Implicit-defaulting the music model
• Persisting tokens, signed URLs, cookies, or session material

Default prompt:

TTS requires --voice-name; music requires --model + --duration. Never pass --count on audio nodes. Never default the music model. Hand off paid execution to dreamina-canvas-quote-and-run.