Synchronous text-to-speech synthesis (≤10,000 chars). Returns audio URL + metadata. Use output_format="url" to receive a 24h-valid URL.

Create an async long-text T2A task (up to 1M chars via text_file_id, or 50k via text). Returns a task_id to poll with speech_t2a_async_query.

Poll an async T2A task created by speech_t2a_async_create. Returns status (success|failed|expired|processing) and a file_id on success.

Clone a voice from an uploaded file_id (purpose=voice_clone). Cloned voices must be used within 168 hours or they are auto-deleted. Returns a demo_audio preview URL.

Generate a video from a text prompt. Returns a task_id — poll with video_agent_query.

Generate a video from an input image (and optional prompt). Returns a task_id — poll with video_agent_query.

Create a video template/agent task. Returns a task_id to poll with video_agent_query.

Poll a video task. Returns status (Preparing|Processing|Success|Fail) and a video_url valid 9h on Success.

Generate images from a text prompt. Returns image URLs (typically valid 24h).

Transform a reference image guided by an optional prompt. Returns image URLs.

Generate music. Provide lyrics for vocal tracks; set is_instrumental=true for instrumental. music-cover* models require audio_url/audio_base64/cover_feature_id.

Upload a file (document, audio, image, etc.) for downstream use. Use purpose=voice_clone before speech_voice_clone.

List uploaded files, optionally filtered by purpose.

Get metadata for a single file by file_id.

Download a file. By default returns a summary (filename + bytes + content-type); pass as_hex=true to inline small file content as a hex string.

Delete a file by file_id.