Docs

Vertex Veo long-running video generation. POST starts an operation; GET polls status.

Videos

POST /v1/videos/generations
GET  /v1/videos/generations/:id
GET  /v1/videos/generations/:id/content
GET  /v1/videos/models

Body: { "model", "prompt", "n", "duration", "size" | "aspect_ratio", "image?" }.

Routes to Vertex Veo (predictLongRunning + fetchPredictOperation) using the same ADC / GOOGLE_CLOUD_PROJECT as chat. Returns 503 if Vertex is not configured. Never invents a video.

X-OpenDoor-Backend: private or model: "premium:private" returns 501 — private GPU video is not implemented.

Live probe on project-800192c2-3ecc-4889-8f7 (us-central1):

Model	Result
`veo-3.1-fast-generate-001`	200 + operation, then `videos[].bytesBase64Encoded`
`veo-3.1-generate-001`	200 + operation, then inline mp4 bytes
`veo-3.0-generate-001` / `veo-3.0-fast-generate-001` / `veo-2.0-generate-001`	404 — Model Garden Enable

duration is seconds (Veo 3: 4, 6, or 8). image is optional base64 / data URL for image-to-video.

Jobs are in-memory, with metadata and mp4 bytes stored in OPENDOOR_FILES_BUCKET when set.

bash
curl http://localhost:3001/v1/videos/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "veo-3.1-fast-generate-001",
    "prompt": "A red cube sitting still on a white table",
    "duration": 4,
    "aspect_ratio": "16:9"
  }'

bash
curl http://localhost:3001/v1/videos/generations/vid_… \
  -H "Authorization: Bearer YOUR_API_KEY"

bash
curl http://localhost:3001/v1/videos/generations/vid_…/content \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -o out.mp4

POST returns { "id", "status": "processing", "model" }. Poll GET until completed (then url is /v1/videos/generations/:id/content) or failed.

Images

Audio