日本語ではまだ利用できません — 英語で表示しています。

`POST /v1/videos/generations`

Generate a short video from a text prompt, optionally conditioned on a first-frame image. Routes to providers running a video service (LTX-2). The response is the raw MP4 bytes — not JSON. Video generation is slow; expect minutes, not seconds.

Request

curl https://api.inference.club/v1/videos/generations \
  -H "Authorization: Bearer $INFERENCE_CLUB_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ltx-2",
    "prompt": "a paper boat drifting down a rain-soaked gutter, cinematic",
    "duration": 6,
    "fps": 24
  }' \
  --output video.mp4

Field	Type	Notes
`model`	string	Required. A model with `service_type: video` from `GET /v1/models`.
`prompt`	string	Required.
`negative_prompt`	string	Optional.
`image`	string	Optional first frame for image-to-video: a `data:` URI or raw base64.
`image_strength`	number	0–1, default 1. How strongly the image constrains the result.
`duration`	number	Seconds, 1–20.
`num_frames` / `fps`	integer	Alternative to `duration`; frames ≤ 1281, fps ≤ 60.
`width` / `height`	integer	64–1920.
`num_inference_steps`	integer	1–100.
`guidance_scale`	number	0–30.
`enhance_prompt`	boolean	Let the server expand your prompt.
`seed`	integer	Optional fixed seed.

Also accepts the visibility and collection sharing fields.

Response

The finished video (video/mp4), with the actual resolved parameters (frames, fps, resolution) recorded on the stored request. Generated videos appear in Dashboard → Watch and can be organized into playlists via collections.

Errors use the standard error format: 404 no_provider when no online provider serves the model, 502 upstream_error if the provider fails mid-render.