Bearer token

Output languages, e.g. ["zh-cn", "en"]. The first is the primary language; each additional one reuses the visuals with a localized voiceover + render and is billed +6 credits.

Optional visual-style preset — expressed through typography, palette, motion, image treatment and pacing. Orthogonal to `scenario` (it does NOT change routing). `auto` (default) lets the director pick; every other value adopts a real named look: `cinematic` = dark film-noir (black + blood-red, Oswald); `glass` = Apple / iOS-26 frosted liquid glass; `luxury` = timeless near-black + indigo, huge whitespace; `swiss` = precise grid + electric blue + oversized numerals; `modern` = clean light SaaS; `editorial` = cream magazine + serif; `warm` = intimate cream + amber; `vibrant` = festive folk colour; `neon` = electric neon glow; `mono` = grayscale, type-led; `pastel` = soft candy pastels; `bold` = huge poster type; `industrial` = raw glitch + rust; `futuristic` = particle glow. A freeform string is also accepted as a soft hint.

generate = a new video. remix / edit / extend = iterate on a previous video (require `ref_task_id`).

Output aspect ratio (hint — the agent may follow the prompt).

Natural-language brief describing the video to produce (the topic, what to show, tone, audience). The agent decides the script, visuals, voiceover and edit.

Production tier, a multiplier on the duration-based price. `draft` = a fast rough cut for previewing the idea (~0.5× the standard credits); `standard` = balanced (default, 1×); `premium` = richer, more detailed and polished (~2× the standard credits). Affects turnaround, detail and price.

Target video length in seconds (1–600, i.e. up to 10 minutes). Billed by duration: credits ≈ (60 + 1.7 × duration) × quality multiplier, so a longer video costs proportionally more.

How to route the video — a hint; the AI director still decides the final structure. `auto` (default) = the director chooses from your brief. `narrated` = multi-scene narrated video with real photos + voiceover + data cards (people / brands / explainers / history / products). `drama` = acted short drama with characters + dialogue (短剧). `avatar` = talking-head / digital human (needs a portrait image via `file_urls`, or a chosen digital human). `motion` = abstract kinetic-typography / data / logo motion graphic. `slideshow` = presentation deck / pitch. Legacy values `general` / `explainer` / `product` / `website` / `changelog` / `captions` are still accepted (mapped to `auto`), and `slides` maps to `slideshow`.

Optional reference media (image / video / audio URLs) the agent can use — e.g. a product shot or logo to feature, footage to caption.

Required when `action` is remix / edit / extend: the task_id of the previous video to start from.

Optional. Fired with the result when the task reaches a terminal state (succeeded / failed).