Generate video and track jobs

Video generation is asynchronous. A successful submission means the job was accepted, not that the final video is ready.

Endpoints

POST /v1/videos
GET  /v1/videos/{task_id}

Text to video

curl -X POST http://localhost:9890/v1/videos \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_GATEWAY_KEY" \
  -d '{
    "model": "agnes-video-v2.0",
    "prompt": "a cinematic orange cat walking through a rainy neon street",
    "width": 1152,
    "height": 768,
    "num_frames": 121,
    "frame_rate": 24
  }'

Image to video

{
  "model": "agnes-video-v2.0",
  "prompt": "slow head turn, subtle background movement, natural motion",
  "image": "https://example.com/input.jpg",
  "width": 1152,
  "height": 768,
  "num_frames": 121,
  "frame_rate": 24
}

Start and end frames

{
  "model": "agnes-video-v2.0",
  "prompt": "smoothly transition from the first frame to the second",
  "images": [
    "https://example.com/frame_start.jpg",
    "https://example.com/frame_end.jpg"
  ],
  "mode": "keyframes",
  "width": 1152,
  "height": 768,
  "num_frames": 121,
  "frame_rate": 24
}

Poll job status

curl -H "Authorization: Bearer YOUR_GATEWAY_KEY" \
  http://localhost:9890/v1/videos/TASK_ID

Poll at a reasonable interval until the job completes or fails. Do not create a high-frequency polling loop.

Common durations

num_frames follows 8n+1 and is capped at 441. Validate with 81 or 121 frames before increasing resolution or duration.