Docs › APIs

Generate videos

Video models run as tasks: you submit a request and get a task ID straight away, check the task until it finishes, then download the video. Every video model on SteadyRouter uses the same endpoints, POST /v1/video/generations to submit and GET /v1/video/generations/{task_id} to check, with your key in Authorization: Bearer.

Pricing

• Video is charged per video. The price depends on the model, the resolution and the length you ask for. Each model's current prices are on Model Square.
• A video costs far more than a chat request. The US$0.05 of credit that new accounts currently start with usually isn't enough for one video. Add funds before you start: see Top-ups and refunds.
• The price is taken from your balance when the task is accepted. If the task fails, it is refunded in full.
• Seedance models (doubao-seedance-*) are charged an estimate when you submit, then settled when the video is ready, based on the tokens the video actually used: you get the difference back, or pay the difference.
• Always set the resolution. If you leave it out, many models use 1080p, which usually costs more.

Before you start

1. Create a SteadyRouter account and add funds.
2. In the console, open API Keys and copy your key. New accounts come with one, Default Key; to make another, click Create API Key.
3. The examples below read the key from an environment variable:
   ShellCopy

   export STEADYROUTER_API_KEY="sk-..."

Step 1: Submit the task

This example makes a 5-second, 540p video from a text prompt with viduq3-turbo:

curl https://api.steadyrouter.com/v1/video/generations \
  -H "Authorization: Bearer $STEADYROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "viduq3-turbo",
    "prompt": "a red apple slowly rotating on a white table, soft light",
    "size": "540p",
    "duration": 5
  }'

The reply comes back right away. Keep the task_id:

{
  "id": "task_4fQ9pX2mLc7RtB1vN8sK3wZ6yH0dJ5aE",
  "task_id": "task_4fQ9pX2mLc7RtB1vN8sK3wZ6yH0dJ5aE",
  "object": "video",
  "model": "viduq3-turbo",
  "status": "queued",
  "progress": 0,
  "created_at": 1793000000
}

Step 2: Check the task

curl https://api.steadyrouter.com/v1/video/generations/task_4fQ9pX2mLc7RtB1vN8sK3wZ6yH0dJ5aE \
  -H "Authorization: Bearer $STEADYROUTER_API_KEY"

Check every 10 to 15 seconds. A video usually takes from one to a few minutes. While it runs, the reply looks like this (some fields left out):

{
  "code": "success",
  "message": "",
  "data": {
    "task_id": "task_4fQ9pX2mLc7RtB1vN8sK3wZ6yH0dJ5aE",
    "status": "IN_PROGRESS",
    "progress": "30%",
    "fail_reason": "",
    "submit_time": 1793000000,
    "start_time": 1793000022,
    "finish_time": 0
  }
}

You can check a task with any key from the same account.

Step 3: Download the video

curl -o video.mp4 \
  https://api.steadyrouter.com/v1/videos/task_4fQ9pX2mLc7RtB1vN8sK3wZ6yH0dJ5aE/content \
  -H "Authorization: Bearer $STEADYROUTER_API_KEY"

This works the same way for every video model; result_url from step 2 also points to the video. Videos are kept for a limited time only, so save yours soon after the task finishes.

Parameters by model

Every model takes model and prompt. The other fields depend on the model family:

• Vidu and Wan: "size" sets the resolution and "duration" the length in seconds, as in step 1.
• Seedance: put both inside "metadata", as "resolution" and "duration" (example below).
• Kling: "mode" is "std" or "pro", and "duration" the length in seconds. "size" sets the shape: "1280x720" (16:9), "720x1280" (9:16) or "1024x1024" (1:1, the default). For kling-video, also set the version: "metadata": {"model_name": "kling-v3"} (the only version offered; example below).
• Veo: "size" is required, "1280x720" or "720x1280"; "seconds", if you set it, is "8". Every Veo video is about 8 seconds long, and each Veo model charges one price per video, whatever the size.
• MiniMax (Hailuo): "duration" is required, in whole seconds, and the resolution goes inside "metadata" as "resolution", written "768P" or "2K" exactly like that. "ratio" (for example "16:9") can go in "metadata" too. A starting image goes in "images", or a first and last frame in "metadata" as "first_frame" and "last_frame"; each image adds a small charge (example below). Video and audio inputs aren't supported.
• A starting image goes in "images": ["https://..."], or in "image": "https://..." for Wan and Kling.

The gateway refuses values outside this table (see the errors below). Write resolutions in lower case, as shown, except 768P and 2K for MiniMax-H3.

Kling (kling-video, kling-video-v4), Veo (veo_3_1, veo_3_1-fast) and MiniMax (MiniMax-H3) are currently offered in groups from a single supplier, so they don't have the Failover tag on Model Square: if that supplier fails, there is no other group to move the request to (see Models and groups).

For example, a 4-second, 480p Seedance video:

{
  "model": "doubao-seedance-2-0-mini-260615",
  "prompt": "a red apple slowly rotating on a white table, soft light",
  "metadata": {"resolution": "480p", "duration": 4}
}

And a 5-second Kling v3 video in standard mode:

{
  "model": "kling-video",
  "prompt": "a red apple slowly rotating on a white table, soft light",
  "mode": "std",
  "duration": 5,
  "size": "1280x720",
  "metadata": {"model_name": "kling-v3"}
}

And a 6-second MiniMax-H3 video at 768P from a starting image:

{
  "model": "MiniMax-H3",
  "prompt": "a red apple slowly rotating on a white table, soft light",
  "duration": 6,
  "images": ["https://example.com/apple.png"],
  "metadata": {"resolution": "768P"}
}

If something goes wrong

When you email support@steadyrouter.com about a video, include its task_id.

More: API docs · Model Square