Wan 3.0 Video Prime

Quick example

curl https://reapi.ai/api/v1/videos/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan3.0-video-prime",
    "prompt": "A lighthouse keeper climbing a spiral staircase at dawn",
    "resolution": "720P",
    "aspect_ratio": "16:9",
    "duration": 10
  }'

import requests

resp = requests.post(
    "https://reapi.ai/api/v1/videos/generations",
    headers={
        "Authorization": "Bearer YOUR_API_KEY",
        "Content-Type": "application/json",
    },
    json={
        "model": "wan3.0-video-prime",
        "prompt": "A lighthouse keeper climbing a spiral staircase at dawn",
        "resolution": "720P",
        "aspect_ratio": "16:9",
        "duration": 10,
    },
)
task_id = resp.json()["id"]

const resp = await fetch('https://reapi.ai/api/v1/videos/generations', {
  method: 'POST',
  headers: {
    Authorization: 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    model: 'wan3.0-video-prime',
    prompt: 'A lighthouse keeper climbing a spiral staircase at dawn',
    resolution: '720P',
    aspect_ratio: '16:9',
    duration: 10,
  }),
});
const { id } = await resp.json();

body := strings.NewReader(`{
  "model": "wan3.0-video-prime",
  "prompt": "A lighthouse keeper climbing a spiral staircase at dawn",
  "resolution": "720P",
  "aspect_ratio": "16:9",
  "duration": 10
}`)

req, _ := http.NewRequest("POST", "https://reapi.ai/api/v1/videos/generations", body)
req.Header.Set("Authorization", "Bearer YOUR_API_KEY")
req.Header.Set("Content-Type", "application/json")
resp, err := http.DefaultClient.Do(req)

Authentication

Every request needs an API key in the Authorization header:

Authorization: Bearer YOUR_API_KEY

Create one in API keys.

Endpoint

POST https://reapi.ai/api/v1/videos/generations

Generation is asynchronous. The response carries a task id; poll GET /api/v1/tasks/{id} until status settles.

Modes

There is no mode field — the model picks the mode from which media fields are present. The two media families are mutually exclusive and a request that mixes them is rejected.

In reference mode, address your material from the prompt as Image1, Video1, Audio1 and so on, numbered by position within each array.

Request body

Content filter

content_filter defaults to true. Requests authenticated with an API key may set it to false; requests originating from the reAPI website are rejected with 20003 if they try, regardless of the value sent.

Response envelope

Submitting returns a task id:

{ "id": "task_01k9s419324drezfbwnsvxyr6h", "status": "processing" }

Poll GET /api/v1/tasks/{id}. A completed task looks like:

{
  "id": "task_01k9s419324drezfbwnsvxyr6h",
  "model": "wan3.0-video-prime",
  "status": "completed",
  "created_at": 1762853430,
  "output": {
    "video_urls": [
      "https://cdn.reapi.ai/media/tasks/01k9s419324drezfbwnsvxyr6h/0.mp4"
    ]
  },
  "usage": { "credits": 1386 },
  "error": null
}

usage.credits is what the task actually cost once settled — see Tasks.

Pricing

Billing has two dimensions: output seconds × resolution.

Neither the mode nor the audio track changes the rate — a document-to-video request costs the same per second as a text-to-video one at the same resolution, with or without sound. Current per-second rates for each resolution are on the model page.

credits = ceil(per_second_usd × markup × duration × 1000)

1 credit = $0.001 USD.

Two things to know before you send a request:

• Omitting resolution bills at 1080P, the most expensive tier, because that is the model's own default.
• duration: -1 reserves the 30-second ceiling. The final length is not knowable at submit time, so the hold is taken at the maximum. Pass an explicit duration when cost matters.

Failed tasks are refunded automatically.

Errors

Full catalog: Errors.

Tips

• Iterate at 480P and switch to 1080P only for the final render — the rate scales with the tier and the framing does not change.
• Reference images hold characters and wardrobe across a long take. Name them positionally in the prompt (Image1, Image2) rather than describing them again in prose.
• With reference video, remember input seconds count against the 30-second budget: a 10-second reference leaves 20 seconds of output.
• Write prompts as a shot list with explicit time ranges. The model follows timed beats more reliably than a single paragraph of description.
• Pick the standard Wan 3.0 tier when turnaround is not the constraint — it takes the identical request body at a lower rate.

Related

• Wan 3.0 — the standard tier of the same model
• Tasks — polling, usage.credits, and task lifecycle
• Errors — full error code catalog