creative-studio-eight-vert.vercel.app

{
  "name": "create_project",
  "kind": "act",
  "impl": "imperative",
  "description": "Starts a new video campaign in the studio. Sets the campaign name, goal, target audience, platform, and visual style. Call this once at the very start of a new campaign, before any other studio tool. When SEED_DEMO=true (env), the studio hydrates from a pre-recorded manifest so the deployed live URL can show real Veo 3 + OpenAI artifacts without burning provider budget per click.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "name": {
        "type": "string",
        "description": "Short, human-readable campaign name."
      },
      "goal": {
        "type": "string",
        "description": "What this video should accomplish, in plain language."
      },
      "audience": {
        "type": "string",
        "description": "Who the video is for."
      },
      "platform": {
        "type": "string",
        "enum": [
          "instagram",
          "youtube",
          "tiktok",
          "linkedin",
          "generic"
        ],
        "description": "Where this video will primarily be published."
      },
      "style": {
        "type": "string",
        "enum": [
          "professional",
          "casual",
          "dramatic",
          "playful",
          "cinematic"
        ],
        "description": "Overall creative tone for the campaign."
      },
      "targetDurationSeconds": {
        "type": "number",
        "description": "Approximate target length of the finished video, in seconds."
      }
    },
    "required": [
      "name",
      "goal",
      "audience",
      "platform",
      "style"
    ]
  }
}

{
  "name": "generate_script",
  "kind": "act",
  "impl": "imperative",
  "description": "Scriptwriter agent: writes a beat-sheet video script (Hook → Pain → Promise → Proof → Objection → Payoff → CTA) as concrete, spoken narration lines powered by the project brief, platform, audience, and style. Each scene's description is a real VO line that feeds directly into Speechify TTS. Call after brand guidelines exist and before any visual or audio generation.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "sceneCount": {
        "type": "number",
        "description": "How many scenes/beats the script should have. Short-form platforms (Instagram, TikTok) work best at 5-7; YouTube and LinkedIn can take 8+."
      },
      "keyMessage": {
        "type": "string",
        "description": "The single most important idea the script must land (e.g. 'drive signups for an eco walking-shoe launch'). Falls back to the project brief goal."
      }
    },
    "required": [
      "sceneCount",
      "keyMessage"
    ]
  }
}

{
  "name": "create_storyboard",
  "kind": "act",
  "impl": "imperative",
  "description": "Graphic Designer agent: turns the existing script scenes into a storyboard by writing a concrete image-generation prompt for each scene. Call after generate_script and before generate_image.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "visualStyleNotes": {
        "type": "string",
        "description": "Art-direction notes to apply across every scene."
      }
    },
    "required": [
      "visualStyleNotes"
    ]
  }
}

{
  "name": "generate_image",
  "kind": "act",
  "impl": "imperative",
  "description": "Graphic Designer agent: generates the still key visual for one scene from its storyboard prompt. Call once per scene after create_storyboard.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "sceneId": {
        "type": "string",
        "description": "The scene id, e.g. 'scene_1'."
      },
      "promptOverride": {
        "type": "string",
        "description": "Optional replacement prompt."
      }
    },
    "required": [
      "sceneId"
    ]
  }
}

{
  "name": "text_to_video",
  "kind": "act",
  "impl": "imperative",
  "description": "Motion Graphics agent: generates a short video clip for a scene from a text prompt.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "sceneId": {
        "type": "string"
      },
      "durationSeconds": {
        "type": "number",
        "description": "Target clip length in seconds (2–10)."
      },
      "motionNotes": {
        "type": "string"
      }
    },
    "required": [
      "sceneId",
      "durationSeconds"
    ]
  }
}

{
  "name": "image_to_video",
  "kind": "act",
  "impl": "imperative",
  "description": "Motion Graphics agent: animates an existing scene key visual into a short video clip.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "sceneId": {
        "type": "string"
      },
      "durationSeconds": {
        "type": "number",
        "description": "Target clip length in seconds (2–10)."
      },
      "motionNotes": {
        "type": "string",
        "description": "Camera/motion direction, e.g. 'slow push in, parallax'."
      }
    },
    "required": [
      "sceneId",
      "durationSeconds"
    ]
  }
}

{
  "name": "text_to_speech",
  "kind": "act",
  "impl": "imperative",
  "description": "Voiceover agent: converts a scene's narration line into spoken audio.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "sceneId": {
        "type": "string"
      },
      "line": {
        "type": "string",
        "description": "The narration line to voice."
      },
      "voiceTone": {
        "type": "string",
        "enum": [
          "warm",
          "energetic",
          "authoritative",
          "calm",
          "playful"
        ],
        "description": "Vocal tone."
      }
    },
    "required": [
      "sceneId",
      "line",
      "voiceTone"
    ]
  }
}

{
  "name": "write_caption",
  "kind": "act",
  "impl": "imperative",
  "description": "Copywriter agent: writes short-form on-screen text or a social caption for one scene.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "sceneId": {
        "type": "string"
      },
      "purpose": {
        "type": "string",
        "enum": [
          "on_screen_text",
          "post_caption",
          "hook_line"
        ],
        "description": "Type of copy."
      }
    },
    "required": [
      "sceneId",
      "purpose"
    ]
  }
}

{
  "name": "compose_video",
  "kind": "act",
  "impl": "imperative",
  "description": "Video Editor agent: assembles all scene clips, narration, and captions into a single timeline.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "transitionStyle": {
        "type": "string",
        "enum": [
          "cut",
          "crossfade",
          "whip_pan",
          "match_cut"
        ],
        "description": "Style of transition between scenes."
      }
    },
    "required": [
      "transitionStyle"
    ]
  }
}

{
  "name": "review_video",
  "kind": "act",
  "impl": "imperative",
  "description": "Critic/QA agent. Does NOT create content. Reviews the composed video and returns APPROVED or NEEDS_REVISION with notes. Read-only.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "checklistNotes": {
        "type": "string"
      }
    }
  }
}

{
  "name": "request_human_approval",
  "kind": "act",
  "impl": "imperative",
  "description": "Pauses the campaign and asks the human to approve or reject before proceeding. The human veto: call it before marking a campaign complete, before anything would be published or sent externally, and any other time a specialist's output needs a human sign-off. Execution does not resume until the human responds. Use sparingly — this is a blocking checkpoint.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "summary": {
        "type": "string",
        "description": "One-line summary of what's being approved."
      },
      "detail": {
        "type": "string",
        "description": "Fuller context the human needs to decide."
      }
    },
    "required": [
      "summary",
      "detail"
    ]
  }
}

{
  "name": "get_project_roadmap",
  "kind": "answer",
  "impl": "imperative",
  "description": "Project Manager agent. Does NOT create or judge content. Returns the full production roadmap and the current phase. Read-only; safe to call at any time.",
  "inputSchema": {
    "type": "object",
    "properties": {}
  }
}

{
  "name": "get_project_status",
  "kind": "answer",
  "impl": "imperative",
  "description": "Project Manager agent. Does NOT create or judge content. Returns a snapshot of the current project: name, phase, per-scene asset completeness, and any open QA notes or pending approvals. Read-only; safe to call at any time, including before deciding what to do next.",
  "inputSchema": {
    "type": "object",
    "properties": {}
  }
}

{
  "name": "refine_scene",
  "kind": "act",
  "impl": "imperative",
  "description": "Refines an existing scene from specific feedback and/or a list of property/value changes. Applied from the Critic/QA NEEDS_REVISION loop, or from a human asking for a targeted edit. Pass sceneId plus optional feedback (why it needs work) and a changes array covering one or more of: description, imagePrompt, caption, durationSeconds. When the scene has a key visual, the still is re-generated against the feedback.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "sceneId": {
        "type": "string",
        "description": "The scene id, e.g. \"scene_2\"."
      },
      "feedback": {
        "type": "string",
        "description": "Why the scene needs refinement — this drives re-generation."
      },
      "changes": {
        "type": "array",
        "description": "Specific property/value edits to apply to the scene.",
        "items": {
          "type": "object",
          "properties": {
            "property": {
              "type": "string",
              "description": "One of: description, imagePrompt, caption, durationSeconds."
            },
            "value": {
              "type": "string",
              "description": "The new value."
            }
          }
        }
      }
    },
    "required": [
      "sceneId"
    ]
  }
}

{
  "name": "export_video",
  "kind": "act",
  "impl": "imperative",
  "description": "Video Editor agent: confirms the composed video is ready to ship and reports what export would produce. With download=true an agent running in the browser triggers the actual file download. Without a real mp4 on disk the studio falls back to a per-scene slideshow manifest that the VideoPreview component stitches in-browser.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "download": {
        "type": "boolean",
        "description": "When true (browser-side only), trigger the actual file download."
      }
    }
  }
}

{
  "name": "list_available_providers",
  "kind": "answer",
  "impl": "imperative",
  "description": "Returns which generation providers are configured for each capability (image, tts, text-to-video, image-to-video, compose). Lets an external agent see whether the studio will call real APIs (OpenAI, fal.ai, Speechify, etc.) or fall back to deterministic placeholders under DEMO_MODE. Read-only.",
  "inputSchema": {
    "type": "object",
    "properties": {}
  }
}

{
  "name": "procedural_video",
  "kind": "act",
  "impl": "imperative",
  "description": "Generates a real video file from a scene's still image by animating it with Ken Burns / parallax / pulse / glide / tilt-shift transforms and recording the result via MediaRecorder. Browser-only fallback used when external video providers (Veo, FAL, Luma, Runway, Replicate) are not reachable. Uploads the WebM to /public/assets/ and stores the public URL on scene.videoUrl so the scene plays as a real video, not a static image.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "sceneId": {
        "type": "string",
        "description": "ID of the scene to generate procedural video for. Must already have an imageUrl."
      },
      "pattern": {
        "type": "string",
        "enum": [
          "kenBurns-in",
          "kenBurns-out",
          "parallax-drift",
          "glide-up",
          "pulse-zoom"
        ],
        "description": "Ken Burns pattern. Default 'kenBurns-in'."
      },
      "logToActivity": {
        "type": "boolean",
        "description": "When true, log this call to the activity feed and update the motion-graphics agent status."
      }
    },
    "required": [
      "sceneId"
    ]
  }
}