DocsAPI reference

OpenAPI specification

Model IDs and options in examples are illustrative. Select an available model from GET /v1/models and check its capabilities and current prices.

Video

Video is generated as a job. POST /v1/videos follows OpenAI’s video API: it checks the options, holds the estimated charge and answers at once with a video object whose status is queued. Poll GET /v1/videos/{id} every few seconds until it is completed, then download the MP4 from GET /v1/videos/{id}/content. Most clips take one to a few minutes.

curl https://ark.dewaweb.com/v1/videos \
  -H "Authorization: Bearer ark_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: teaser-001" \
  -d '{"model":"google/veo-3.1-lite","prompt":"A slow drone shot over misty rice terraces at sunrise","seconds":4,"resolution":"720p","aspect_ratio":"16:9"}'

curl https://ark.dewaweb.com/v1/videos/VIDEO_ID -H "Authorization: Bearer ark_live_YOUR_KEY"
curl https://ark.dewaweb.com/v1/videos/VIDEO_ID/content -H "Authorization: Bearer ark_live_YOUR_KEY" --output video.mp4
ParameterTypeDefaultDescription
modelstringrequiredAn available video model. Its video metadata lists durations, resolutions, aspect ratios and audio support.
promptstringrequiredWhat to film, up to the model’s video.max_prompt_chars.
secondsinteger or stringmodel defaultOne of the model’s video.durations.
resolutionstringmodel default480p, 720p, 1080p or 4k, from video.resolutions.
aspect_ratiostringfirst listedFrom video.aspect_ratios, for example 16:9 or 9:16.
sizestringoptionalAlternative to resolution and aspect_ratio, as listed in video.sizes, for example 1280x720.
generate_audiobooleantrue where supportedfalse for a silent clip where video.silent is true. Silent clips cost less on some models.
seedintegeroptionalWhere video.seed is true.
negative_promptstringoptionalWhere video.negative_prompt is true.

Statuses are queued, in_progress, completed, failed and cancelled. A failed job carries an error with a code such as content_filtered or generation_timeout. GET /v1/videos lists your jobs, newest first, with limit and after. The download supports byte ranges, so browsers and players can seek. Videos are kept for 7 days.

DELETE /v1/videos/{id} cancels a job that has not reached the provider yet, releases its hold and answers 200. While the job is being submitted it answers 202: the cancellation is recorded, so poll the job to see whether it took. Once the provider has the job, Ark cancels it only if the provider still has it queued and confirms; otherwise it answers 409 job_not_cancellable, and the job runs to the end and is charged. On a finished job it deletes the stored video.

The hold covers the requested length at the model’s price. The charge settles once the video is stored: per second of video for Veo, by output tokens for Seedance. A job that fails, is refused, times out or is cancelled is not charged. Reusing an Idempotency-Key with the same request returns the same job in its current state instead of starting another, even when that job failed or was cancelled; send a new key to try again.