API

Video generation

Video generation is asynchronous: create a job, poll it until it finishes, then download the file.

Create a job

curl https://api.llm-hub.dev/v1/videos \
  -H "Authorization: Bearer $LLMHUB_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google-flow-veo-fast",
    "prompt": "A banana-shaped spacecraft crossing Saturn rings",
    "duration_seconds": 8,
    "aspect_ratio": "16:9",
    "resolution": "720p",
    "client_request_id": "order-42-video-1"
  }'

The API returns 202 Accepted with a job such as {"id": "vidjob_...", "object": "video.job", "status": "queued", "cost_usd": ...}. client_request_id (or an Idempotency-Key header) makes retries safe: repeating it returns the same job.

Poll, download, cancel

curl https://api.llm-hub.dev/v1/videos/vidjob_JOB_ID \
  -H "Authorization: Bearer $LLMHUB_API_KEY"

curl https://api.llm-hub.dev/v1/videos/vidjob_JOB_ID/content \
  -H "Authorization: Bearer $LLMHUB_API_KEY" --output result.mp4

curl -X DELETE https://api.llm-hub.dev/v1/videos/vidjob_JOB_ID \
  -H "Authorization: Bearer $LLMHUB_API_KEY"

status moves through queued, running, and ends as succeeded, failed, or canceled. A succeeded job includes content_url. Failed and canceled jobs are refunded to the key balance.

Last updated 2026-09-30