Generate Videos with Grok

What you can do

Use WorldRouter’s async video task API to generate videos with xAI Grok. The flow is simple: create a task, poll until it reaches a terminal state, then download the output video from the task response.

Before you start

• You need a team-scoped API key with available credits.
• Use the same base URL from Quickstart. Grok video lives under the OpenAI-compatible /v1 surface.
• Grok video requests are asynchronous. A successful create request returns a task id; the video is only ready after the task status becomes succeeded.

Supported models

These are the request model names you send in the model field:

Grok accepts a resolution of 480p or 720p, defaulting to 720p. Any other value — or an omitted one — falls back to 720p rather than returning an error.

Quick start flow

1. 1Create a video task with POST /videos/generations.
2. 2Save the returned task id.
3. 3Poll GET /videos/generations/{task_id} until the task reaches a terminal state.
4. 4When status becomes succeeded, download the video from content.video_url.

Create a task

Grok request bodies use lowercase fields. WorldRouter forwards these requests to xAI’s video surface through the async task API.

Use grok-imagine-video for standard text-to-video generation:

curl -X POST "https://inference-api-pre-d80ca3.worldrouter.ai/v1/videos/generations" \
-H "Authorization: Bearer your_api_key" \
-H "Content-Type: application/json" \
-d '{
  "model": "grok-imagine-video",
  "prompt": "A drone flyover of a snowy mountain range at sunrise.",
  "duration": 6,
  "resolution": "720p"
}'

Typical create response:

{
"id": "task_123",
"request_id": "task_123",
"model": "grok-imagine-video",
"status": "queued"
}

Use grok-imagine-video-1.5-preview when you want to start from an input image:

{
"model": "grok-imagine-video-1.5-preview",
"prompt": "Pan slowly across the scene.",
"resolution": "720p",
"image": {
  "url": "https://your-host.example.com/start.png"
}
}

Check task status

Poll the task until it reaches a terminal state. Use the task id returned by the create response:

curl "https://inference-api-pre-d80ca3.worldrouter.ai/v1/videos/generations/task_123" \
-H "Authorization: Bearer your_api_key"

Typical status response:

{
"id": "task_123",
"request_id": "task_123",
"model": "grok-imagine-video",
"status": "succeeded",
"content": {
  "video_url": "https://media.example.com/output.mp4"
}
}

WorldRouter normalizes task states into:

• queued
• running
• succeeded
• failed
• cancelled
• expired

Download the output

Only download the result after the task status becomes succeeded:

curl -L "https://media.example.com/output.mp4" -o output.mp4

Read the final video URL from content.video_url in the task status response, not from the create response.

Notes and limits

• Billing is per generated second and settles when the task reaches a terminal state. For Grok, pricing varies by resolution, and image-guided generation may add an image fee. Video billing does not behave like standard text-token billing, so token-count fields may be empty or not useful for billing even when Credits change.
• A task is rejected at create time if your team balance is too low.
• The number of in-flight, or pending, tasks per user is capped. Exceeding the cap returns 429.
• Task reads are owner-scoped: you can only poll a task created by the same authorized account, team, or API key. Cross-owner reads return 403 media_resource_access_denied.
• List routes are disabled. GET /v1/videos/generations without a task id is not supported.

Troubleshooting

See also

• Kling video guide: the Kling async video API.
• Seedance video guide: the Seedance async task API.
• Quickstart: API key and base URL setup.
• Models: current model catalog and pricing.
• API reference: chat-completions docs for text models.