Director — Clip Gen

Generate a video clip for each scene from the start frame (and optionally end frame) produced by /director-frame-gen.

Models

Use whichever model the user configured via /model-provider. If none specified, ask.

Kling 3.0 (via KIE API)

POST https://api.kie.ai/api/v1/jobs/createTask
{
  "model": "kling-3.0/video",
  "input": {
    "prompt": "<character.promptBase>, <scene.prompt>",
    "negative_prompt": "smooth plastic skin, airbrushed skin, beauty filter, floating limbs, disconnected body parts, distorted hands, extra fingers, morphing clothes",
    "image_urls": ["<startFrameUrl>", "<endFrameUrl (if exists)>"],
    "sound": <true or false based on format config>,
    "duration": "10",
    "aspect_ratio": "9:16",
    "mode": "std",
    "multi_shots": false
  }
}

Poll GET /jobs/recordInfo?taskId=<taskId> every 15s until state is "success". Parse resultJson for resultUrls[0].

Sora 2 / Seedance 2

Refer to the provider's API documentation for endpoint details. The pipeline is the same — send a prompt and reference image, get a clip URL back. The QA gates in /director-qa work identically regardless of which model generated the clip.

Output

For each scene, produce a video clip URL. Pass to /director-qa for QA.