Video

Video is generated asynchronously: you create a job, poll its status and download the finished video. It works only with video models from the catalog; in the examples below put their ID in place of VIDEO_MODEL_ID.

1. Create a job#

curl https://ai-seller.vibe-codes.ru/v1/videos \
  -H "Authorization: Bearer $AISELLER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "VIDEO_MODEL_ID",
    "prompt": "Waves crashing against a lighthouse at sunset",
    "seconds": 8
  }'

The response is a 202 with the job object:

{"id": "vid_...", "object": "video", "model": "VIDEO_MODEL_ID", "status": "queued", "progress": 0, "created_at": 1790000000}

2. Poll the status#

curl https://ai-seller.vibe-codes.ru/v1/videos/vid_... \
  -H "Authorization: Bearer $AISELLER_API_KEY"

Poll every few seconds until status becomes completed or failed. The intermediate statuses are queued and in_progress.

{"id": "vid_...", "object": "video", "status": "completed", "seconds": "8", "data": [{"url": "https://.../v1/videos/vid_.../content"}]}

A failed job has status: "failed" and error.code: "video_generation_failed"; it is not billed.

3. Download the video#

The link in data[0].url points to GET /v1/videos/{id}/content. Download it with the same key:

curl -L https://ai-seller.vibe-codes.ru/v1/videos/vid_.../content \
  -H "Authorization: Bearer $AISELLER_API_KEY" \
  -o video.mp4

Before the job completes, this address returns 404 not_found.

Specifics#

• Billing is for the video length in seconds from the provider's response (the requested seconds if it did not report it), once, when the job completes. No balance reserve is held while the video is generated.
• Parameters: prompt, seconds (or duration, duration_seconds), size, aspect_ratio, resolution, negative_prompt, seed; references — image, images, video, input_reference. The request body limit is 70 MiB.
• A job is visible only to keys of your account; someone else's id returns 404 not_found.
• Video jobs are not subject to the concurrent request limit.