plan_generation

Management

Plan a Generation Pipeline

Plan a multi-step generation pipeline for complex media requests (e.g. "product video with voiceover", "concept → 3D asset"). Args: - description: natural-language description of the desired workflow Returns: a matched pipeline with ordered steps, each referencing a tool + optional model hint. Don't use when: You only need one generation step. Prefer recommend for a single model.

read-onlyopen-world

Parameters

Name	Type	Required	Description
description	string	✓	Natural-language description of the multi-step workflow to plan (e.g. 'product launch video with voiceover').
response_format	enum(json \| markdown)	json	Output format: 'json' for structured data, 'markdown' for human-readable text.

Example Request

JSON

{
  "description": "product launch video with voiceover from a hero image"
}

Example Response

JSON

{
  "matched": {
    "key": "product_video",
    "description": "Product image → hero video → voiceover → combine",
    "confidence": 0.85,
    "steps": [
      {
        "tool": "run_model",
        "purpose": "Generate hero product image",
        "model_hint": "model_google-gemini-3-1-flash"
      },
      {
        "tool": "analyze",
        "action": "remove_background",
        "purpose": "Clean transparent product image"
      },
      {
        "tool": "run_model",
        "purpose": "Image-to-video product reveal",
        "model_hint": "model_kling-v3-i2v-pro"
      },
      {
        "tool": "run_model",
        "purpose": "Generate voiceover",
        "model_hint": "model_xai-grok-tts",
        "optional": true
      }
    ]
  },
  "alternatives": []
}

Common Use Cases

Plan a multi-step generation pipeline before kicking off the first run_model call
Discover the standard sequence for common workflows (product video, talking head, game asset)
Get model hints per stage that the host LLM can confirm with the user before executing