Video

Video generation is asynchronous because provider jobs can take from seconds to several minutes. Submit a job, poll its status (or use a webhook), then download the completed MP4 through the gateway.

Workflow

curl -X POST "https://api.llmapi.ai/v1/videos" \
  -H "Authorization: Bearer $LLM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kling-v2.1-master-t2v",
    "prompt": "An eagle flying over snowy mountains at sunrise",
    "seconds": 5
  }'

The 202 Accepted response contains a job id. Poll it until status is completed or failed, then download the content:

curl "https://api.llmapi.ai/v1/videos/VIDEO_ID" \
  -H "Authorization: Bearer $LLM_API_KEY"

curl "https://api.llmapi.ai/v1/videos/VIDEO_ID/content" \
  -H "Authorization: Bearer $LLM_API_KEY" \
  --output video.mp4

The status response is OpenAI Sora-compatible and includes the requested model, duration, size, progress, usage, and—after completion—the output dimensions and provider URL:

{
  "id": "abc123",
  "object": "video",
  "status": "completed",
  "progress": 100,
  "model": "kling-v2.1-master-t2v",
  "seconds": "5",
  "size": "1920x1080",
  "usage": { "duration_seconds": 5 },
  "format": "mp4",
  "duration_seconds": 5,
  "width": 1920,
  "height": 1080
}

Create request fields

Field	Description
`model`	Video model ID. Use the catalog for supported operations and limits.
`prompt`	Text description of the scene and motion. Required by text-to-video and most image-to-video models.
`seconds`	Requested duration as a number or string, for example `5` or `"5"`.
`size`	Resolution such as `1280x720`; `width` and `height` can be used where supported.
`fps`	Requested frame rate, when supported.
`image_url`	Public source image for first-frame image-to-video.
`image_list`	Multiple reference images or start/end frames.
`video_url` / `video_list`	Reference or source videos for supported models.
`webhook`	HTTPS endpoint to receive the completed status payload.
`seed`, `quality`, `generate_audio`	Optional model-specific controls.

For OpenAI-style input references, input_reference: { "image_url": "..." } is accepted where supported. file_id references from the Files API are not implemented for video generation; media URLs must be reachable by the provider.

For image-to-video or reference-to-video models, provide the model-supported image_url, image_list, video_url, or video_list fields. You can provide webhook in the create request instead of polling.

Notes

The /content route is available only after the job completes and streams the MP4 through LLM API.
Polling is usually appropriate every 5–15 seconds for interactive jobs; do not assume that a job completes within a fixed time because provider queues vary.
Media URLs supplied to providers must be reachable by the selected provider.
Webhook delivery is at-least-once and payloads are unsigned; verify the job ID against an authenticated record and make receivers idempotent.
The direct video_url in a completed status is provider-hosted and may expire. Prefer downloading through /v1/videos/{id}/content when the file must be retained.
A failed job includes an error object when available. Common causes are unsupported model parameters, inaccessible media, provider moderation, insufficient credits, or an upstream timeout.
See the full Video Generation guide for model-specific inputs, durations, resolutions, and pricing. Provider fallback also applies to supported video jobs.