Generate designed speech
Generate speech using a natural language voice description.
Describe the desired voice in instruct_text (e.g. 'a deep male voice
with an American accent, speaking confidently'). The model creates a
new voice matching the description each time — results vary between calls.
To reuse a designed voice, save it via POST /api/v1/voices and use
POST /api/v1/generation/design/saved instead.
sync?SyncWait for generation to complete and return audio directly
falseapplication/json- body
instruct_text*Instruct TextNatural language description of the voice
target_text?Target TextText to generate speech for. Required for most models; for Dramabox, the instruct_text prompt contains both stage directions and spoken text.
""language?LanguageTarget language. Query GET /api/v1/models/active for supported languages per variant.
"Auto"generation_params?|Advanced sampling parameters
model_family?|TTS model to use. If omitted, uses the currently configured model.
voice_reference_path?|Path to a reference audio file for voice conditioning (Dramabox only).
Generation started (async) or completed result (sync).
application/json- response
started*StartedWhether the generation job was enqueued
queue_position*Queue PositionPosition in the generation queue (0 = next to run)
curl -X POST 'http://127.0.0.1:8100/api/v1/generation/design' \ -H 'Content-Type: application/json' \ -d '{ "instruct_text": "A warm female voice with a British accent, speaking at a moderate pace"}'{ "started": true, "queue_position": 0}Stream cloned speech POST
Start a streaming voice clone session. Returns a `session_id` immediately. Connect to `GET /api/v1/streaming/{session_id}/audio` to receive chunked WAV audio as it is generated. Call `POST /api/v1/streaming/{session_id}/finalize` when done to get the complete audio as base64. Cancels any in-progress generation to prioritize the stream.
Generate from saved design voice POST
Generate speech using a previously saved voice design. Validates voice_id exists on disk.