OmniPort API documentation

Quickstart#

You need an API key in the form sk-op-…. Create one in the console — it is shown once, and only its hash is stored. Your account holds the prepaid balance shared by all its keys; a key only sees its own jobs. Base URL for this deployment:

https://www.omniportapi.com/v1

Create a job, then poll it by id:

curl https://www.omniportapi.com/api/v1/videos \
  -H "Authorization: Bearer $OMNIPORT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sora-2",
    "prompt": "a paper boat drifting through a neon city",
    "size": "720x1280",
    "seconds": 8
  }'

curl https://www.omniportapi.com/api/v1/videos/video_9f2c… \
  -H "Authorization: Bearer $OMNIPORT_API_KEY"

Authentication#

Every request carries one header. Missing, malformed or revoked keys get 401 authentication_error with code: invalid_api_key.

Authorization: Bearer sk-op-...

Credits are held by your account, so rotating a key never moves the balance. Jobs created with a key are only visible to that key — asking for another key's job returns 404 not_found_error, not 403, so job ids cannot be probed for validity.

OpenAI-compatible surface#

Use this surface if you already have OpenAI client code. Model names, size and seconds are translated to the generation pipeline internally.

Create a video

POST /api/v1/videos — JSON, or multipart/form-data when input_reference is a file.

Response — 200 with the Video object:

{
  "id": "video_9f2c…",
  "object": "video",
  "created_at": 1767225600,
  "status": "queued",
  "completed_at": null,
  "error": null,
  "expires_at": null,
  "model": "sora-2",
  "progress": 0,
  "prompt": "a paper boat drifting through a neon city",
  "remixed_from_video_id": null,
  "seconds": "8",
  "size": "720x1280"
}

Statuses

List, fetch, download

Native surface#

Native endpoints take the generation parameters directly instead of a model name. The scenario is chosen by the path — there is no field for it, and sending an unknown field is an error rather than a silent no-op.

Request fields

First / last frame — fl2v

curl https://www.omniportapi.com/api/v1/fl2v/generations \
  -H "Authorization: Bearer $OMNIPORT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "the boat drifts forward, camera slowly pushes in",
    "aspectRatio": "9:16",
    "resolution": "1080P",
    "durationSeconds": 8,
    "imageUrls": ["https://cdn.example.com/frame.png"]
  }'

Upscale — one video in, one higher tier out

curl https://www.omniportapi.com/api/v1/upscale/generations \
  -H "Authorization: Bearer $OMNIPORT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "aspectRatio": "16:9",
    "resolution": "2K",
    "videoUrls": ["https://cdn.example.com/clip.mp4"]
  }'

A typical loop is: validate the prompt at 480P, then upscale the clip you liked to 2K, instead of paying for two full renders.

Accepted request

201 — the job id is the upstream task id, so polling can start immediately:

{ "id": "task_8c41…", "status": "queued", "scenario": "fl2v" }

Task object

GET /api/v1/generations/{id}

{
  "id": "task_8c41…",
  "status": "succeeded",
  "error": null,
  "scenario": "fl2v",
  "resolution": "1080P",
  "aspectRatio": "9:16",
  "durationSeconds": 8,
  "createdAt": 1767225600,
  "updatedAt": 1767225900,
  "artifacts": [{ "kind": "video", "url": "https://…/out.mp4", "filename": "out.mp4" }]
}

Native statuses are passed through as the pipeline reports them: queued, scheduling, running, succeeded, failed, cancelled. artifacts is only filled for succeeded jobs, and is only present on the detail call.

List tasks

GET /api/v1/generations?page_num=1&page_size=20 returns { tasks, total, page_num, page_size }. Page size is capped at 100. The list is newest first and does not include artifacts — fetching artifacts for every row would make the list as slow as the slowest job.

Pricing and billing#

Prices are quoted in credits (1 credit = 1 US cent) per second of generated video, with a minimum per job. The current numbers are on the pricing section of the overview page; they come from this deployment's price list, not from a constant in the docs.

• The full amount in credits is held when the job is created. If your account balance is short, the request fails with 402 insufficient_quota and no job is queued.
• OpenAI surface: billed per model. Native surface: billed per resolution tier. The two price lists are maintained separately.
• The pipeline renders at least 5 seconds, so seconds: 4 is billed as 5.
• Failed or cancelled jobs are refunded automatically. The refund is recorded per job and happens once, no matter how often the job is polled.
• Upscale is billed as one 5-second unit; the real duration comes from the source video.

Buying credits#

Credits are prepaid balance on your account, and every key under that account spends from the same balance. Packs are priced in US dollars and credited as credits (1 credit = 1 US cent); the list on the overview page is the same one shown in the console — one table, so anything you see in one place is purchasable in the other.

• Payment is handled by Paddle. A pack marked unavailable on this deployment has no price configured for it — contact support.
• The balance is credited only when Paddle confirms the transaction. Returning from the checkout page is not a confirmation.
• Credits never expire. Packs with a bonus credit more than they charge, and the bonus is already included in the amount you receive.
• Topping up credits the account, not a key: rotating keys does not move or lose the balance.

Errors#

Every failure, including 5xx, is the same envelope. param and code are optional.

{
  "error": {
    "message": "resolution must be one of 480P, 768P, 1080P, 2K.",
    "type": "invalid_request_error",
    "param": "resolution",
    "code": "invalid_value"
  }
}

Limits#

• input_reference upload: 10 MB, PNG / JPEG / WebP only.
• Prompt: 8000 characters per request.
• Reference media: at most 8 images, 3 videos and 3 audio clips per r2v request; 2 images per fl2v request.
• Reference URLs must be public http(s). Private addresses are refused — the pipeline downloads them, and it cannot reach your intranet.
• List endpoints: at most 100 items per page.