Skip to main content
duration truncates from the start of the input clip. Allowed range is [2, 10]; pick a value ≤ input video length or upstream rejects the job with “failed to exec task”. The upstream’s “no truncation” sentinel (duration: 0) is intentionally disabled on RouterBase — it billed $0 against an arbitrary-length output, so we pin duration to an explicit value.

Endpoint

Async — returns {id, status: "pending"}; poll GET /v1/videos/generations/{id} until success or failed.

Parameters

string
required
Text instruction describing the edit. Max 5000 characters.
string[]
required
Exactly one input-video URL (HTTPS, MP4).
string[]
Up to 3 reference image URLs to bias the edit.
"720P" | "1080P"
default:"\"1080P\""
Output resolution. Drives the per-second rate (0.10/svs0.10/s vs 0.15/s).
integer
default:"5"
Length of the generated clip in seconds. Range [2, 10]. Truncates from the start of the input. Must not exceed the input video’s length or upstream will reject the job.
"16:9" | "9:16" | "1:1" | "4:3" | "3:4"
Output aspect ratio. If omitted, the input clip’s aspect ratio is used.
"auto" | "origin"
default:"\"auto\""
auto lets the model synthesize a new audio track; origin keeps the input clip’s audio.
string
Content to steer the model away from. Max 500 characters.
boolean
default:"true"
Whether the model should expand the prompt before generation.
boolean
default:"false"
Whether to overlay an upstream watermark on the output.
integer
Random-generator initialiser. Range [0, 2147483647].

Example

Poll until terminal:

Reference