每一项能力。
一个余额。

Analyse beat structure of an audio file. Synchronous — returns the beat list directly.

Detect shot boundaries via TransNetV2. Synchronous — returns the shot list directly.

Download social-media media (TikTok / Instagram / Twitter / Bilibili / …) and re-host on CDN.

Fetch a URL and emit extracted text / markdown / raw body (--mode). Non-2xx → error.

Deprecated ambiguous image generation command. Use generate-image-prompt or generate-image-ref.

Generate a still image from a text prompt only; reference image inputs are rejected.

Generate or edit a still image from required source/reference image URL(s).

Generate music from a required prompt; prompt is music direction unless --custom makes it lyrics.

Generate a short sound effect (SFX / Foley / ambient / short loop) from a text description. ElevenLabs Sound Effects v2, duration 0.5–30s.

Generate speech audio from text. `--provider` picks the backend (minimax / elevenlabs / elevenlabs-dialogue / voxcpm / seed-audio).

Generate a short video clip from a text prompt (optionally seeded with reference images / audio).

Forward (address → coords) or reverse (coords → address) geocoding.

Generate a talking-head video where a character's lips sync to a driving audio track. Submits one job (or N via --jobs-file) and returns {job_ref, status:'pending'}; the rendered video URL arrives via [async-callback].

Transfer motion / pose from a source video onto one or more target characters (services: dreamactor, kling3.0-motion). dreamactor supports multi-person via --image-urls; kling3.0-motion is single-person only.

Search places / POIs near coordinates and fetch their photos.

Burn SRT captions into a video. Synchronous — returns the captioned video URL directly.

External web search via Baidu / Serper / Tavily / Brave, plus Baidu similar-image search.

Bird's-eye view map image at coordinates with optional markers.

Street-level imagery and coverage info via Google Maps.

Asynchronously transcribe audio to SRT subtitles. Returns {status:'pending', job_ref} immediately; the SRT lands via [async-callback] in a later turn.