Music Video Technical Script Generator Agent (TypeScript)

Description

The Music Video Technical Script Generator Agent automates the process of creating complex, production-ready scripts for music videos. Drawing on AI-powered text generation, the agent:

1. Generates a script outline (scene breakdowns, camera movements, lighting, etc.).
2. Extracts scene data (start/end times, shot type, transitions).
3. Identifies the settings or environments in each scene with rich descriptors.
4. Extracts all characters mentioned (including musicians, background dancers, extras).
5. Transforms the final scene data into a set of prompts suitable for subsequent AI-based image/video generation.

All steps are event-driven, with tasks managed and billed through Nevermined. By subscribing to step-updated events, this agent listens for instructions and updates each step’s status as it progresses (e.g., from Pending to Completed).

Related Projects

This Movie Script Generator Agent is part of a larger ecosystem of AI-driven media creation. For a complete view of how multiple agents work together, see:

1. Music Orchestrator Agent⁠

   • Coordinates end-to-end workflows: collects user prompts, splits them into tasks, pays agents in multiple tokens, merges final output.

2. Song Generator Agent⁠

   • Produces lyrics, titles, and final audio tracks using LangChain + OpenAI and a chosen music generation API.

3. Image / Video Generator Agent⁠

   • Produces Images / Video using 3rd party wrapper APIs (Fal.ai and TTapi, wrapping Flux and Kling.ai)

Workflow Example:

[ User Prompt ] --> [Music Orchestrator] --> [Song Generation] --> [Script Generation] --> [Image/Video Generation] --> [Final Compilation]

Table of Contents

1. Features
2. Prerequisites
3. Installation
4. Environment Variables
5. Project Structure
6. Architecture & Workflow
7. Usage
8. Detailed Guide: Creating & Managing Tasks
9. Development & Testing
10. License

Features

• Nevermined Integration: Subscribes to tasks via step-updated events, updates the workflow’s steps, and logs progress remotely.
• LangChain + OpenAI: Uses advanced prompt templates to produce detailed cinematic descriptions.
• Scene Extraction & Settings: Splits an existing script into discrete scenes with recommended camera gear, lighting setups, and transitions. Also identifies distinct locations/settings.
• Character Extraction: Provides names, descriptions, wardrobe details, and AI-friendly prompts for each character.
• Transform Scenes: Creates final prompts summarizing each scene in a single JSON array (duration, setting, character references).
• Modular Design: Extend or replace steps with minimal disruption.
• Logging & Error Handling: Comprehensive logs at every step, with robust fallback in case of failures.

Prerequisites

• Node.js (>= 18.0.0 recommended)
• TypeScript (project developed on ^5.x)
• Nevermined credentials (API key, environment config, AGENT_DID)
• OpenAI API Key for text generation

Installation

1. Clone this repository:

   git clone https://github.com/nevermined-io/music-video-script-generator-agent.git
   cd music-video-script-generator-agent
   

   Copy

2. Install dependencies:

   npm install
   

   Copy

3. Build the project (optional for production):

   npm run build
   

   Copy

Environment Variables

Rename .env.example to .env and set the required keys:

NVM_API_KEY=your_nevermined_api_key
NVM_ENVIRONMENT=testing
AGENT_DID=did:nv:your_agent_did
OPENAI_API_KEY=your_openai_api_key
IS_DUMMY=false

• NVM_API_KEY and NVM_ENVIRONMENT configure the connection to Nevermined.
• AGENT_DID sets the unique identifier for this agent.
• OPENAI_API_KEY grants access to OpenAI’s text generation models.
• IS_DUMMY controls whether to use a “dummy” mode (returning static data instead of calling real services).

Project Structure

music-video-script-generator-agent/
├── main.ts                      # Main entry (initializes Payments, subscribes to steps)
├── config/
│   └── env.ts                   # Environment configuration
├── steps/
│   └── stepHandlers.ts          # Core logic for handling each step
├── sceneTechnicalExtractor.ts   # Class for script creation, scene extraction, etc.
├── types.ts                     # Shared TypeScript interfaces (e.g., Scene)
├── logger/
│   └── logger.ts               # Logging system (local + remote)
├── .env.example                 # Environment template
├── package.json
├── tsconfig.json
└── README.md                    # This file

Key Components:

1. main.ts: Handles task lifecycle, from receiving steps to sending back results.
2. sceneTechnicalExtractor.ts: Implements the multi-step logic for generating the initial script, extracting scenes, identifying settings, and extracting characters.
3. stepHandlers.ts: Coordinates the different steps (init, generateScript, extractScenes, generateSettings, extractCharacters, and transformScenes) that shape the final output.
4. logger/logger.ts: A logging system that logs both locally and through the Nevermined Payments API.

Architecture & Workflow

1. Initialization (init step)

   • When a new task is created for this agent, it starts with init.
   • We create subsequent steps (e.g., generateScript, extractScenes, generateSettings, extractCharacters, transformScenes).

2. Script Generation (generateScript)

   • Takes the user’s “idea” (plus optional lyrics/tags/duration) and generates a full technical script.
   • The script typically includes scene descriptions, durations, camera gear, transitions, and other cinematic details.

3. Scene Extraction (extractScenes)

   • Parses the newly generated script to produce structured data: start/end times for each scene, shot types, transitions, etc.

4. Settings Generation (generateSettings)

   • Identifies each unique environment (e.g., rooftop party, city street, beach) and creates a JSON array describing them.

5. Characters Extraction (extractCharacters)

   • Finds all references to characters (lead singer, extras, musicians) in the script.
   • Produces a list with each character’s physical descriptors, wardrobe details, movement style, etc.

6. Scene Transformation (transformScenes)

   • Merges all extracted data to produce final “prompts,” each describing a scene’s visuals, camera movements, participating characters, and links to the correct setting.
   • These prompts can be used directly for downstream AI-based image or video generation tasks.

Throughout these steps, the agent updates each step’s status (from Pending to Completed or Failed) in the Nevermined system. If a step fails, it logs the error and halts.

Usage

After installing and configuring .env:

npm start

1. The agent logs into Nevermined with your NVM_API_KEY and listens for step-updated events for AGENT_DID.
2. When an orchestrator (or higher-level process) triggers a new “init” step for this agent, the agent spawns the subsequent steps and processes them one by one.
3. The final step yields a JSON with scene prompts, character data, and settings.

Detailed Guide: Creating & Managing Tasks

Introduction

The Music Video Technical Script Generator leverages LangChain and OpenAI to create production-ready scripts with precise technical specs—scene-by-scene breakdowns, camera movements, lighting setups, color palettes, lens/stabilizer recommendations, visual references, transitions, and more. All of this is integrated with Nevermined's Payments API for structured task handling and billing.

Installation & Running the Agent

1. Clone and install as described above (npm install, configure .env).
2. Build (optional) with npm run build.
3. Start via npm start. The agent will connect to Nevermined and await new tasks.

Project Structure (Extended)

music-video-script-generator-agent/
├── src/
│   ├── main.ts                       # Agent entry point for step subscription
│   ├── steps/
│   │   └── stepHandlers.ts           # Detailed step logic (init, generateScript, etc.)
│   ├── sceneTechnicalExtractor.ts    # Core logic for generating / extracting script data
│   ├── logger/
│   │   └── logger.ts                 # Logging system
│   └── config/
│       └── env.ts                    # Environment variable load
├── .env.example                      # Environment template
├── package.json                      # Dependencies
├── tsconfig.json                     # TypeScript config
└── README.md

Notable Files

• sceneTechnicalExtractor.ts: Contains the logic to generate an initial script (via OpenAI), then parse out scenes, settings, and characters, returning structured JSON data.
• stepHandlers.ts: Defines the workflow steps, updating the task in the Nevermined system accordingly.

Integration with Nevermined Payments API

Below is a step-by-step outline of how Nevermined orchestrates tasks in this agent:

1. Initialize the Payments Instance

   import { Payments, EnvironmentName } from "@nevermined-io/payments";
   
   const payments = Payments.getInstance({
     nvmApiKey: process.env.NVM_API_KEY!,
     environment: process.env.NVM_ENVIRONMENT as EnvironmentName,
   });
   
   if (!payments.isLoggedIn) {
     throw new Error("Failed to authenticate with Nevermined.");
   }
   

   Copy

   This connects to Nevermined using the API key and environment.

2. Subscribe to Task Updates

   await payments.query.subscribe(run, {
     joinAccountRoom: false,
     joinAgentRooms: [process.env.AGENT_DID!],
     subscribeEventTypes: ["step-updated"],
     getPendingEventsOnSubscribe: false,
   });
   

   Copy

   Here, run is a callback that processes each step-updated event.

3. Task Lifecycle

   • Fetch a step’s details using payments.query.getStep(stepId).
   • Process the step (e.g., generate a script or extract scenes).
   • Update the step in the system using payments.query.updateStep(step.did, { ... }).

1. Subscribing to Task Requests

In main.ts, you’ll typically see:

await payments.query.subscribe(processSteps(payments), {
  joinAccountRoom: false,
  joinAgentRooms: [AGENT_DID],
  subscribeEventTypes: ["step-updated"],
  getPendingEventsOnSubscribe: false,
});

• Event Subscription: The agent will now receive an event each time a relevant step is updated for its DID.
• Callback Execution: The function processSteps(payments) is invoked with the event data.

2. Handling Task Lifecycle

In stepHandlers.ts, you might see something like:

export function processSteps(payments: Payments) {
  return async (data: any) => {
    const eventData = JSON.parse(data);
    const step = await payments.query.getStep(eventData.step_id);

    if (step.name === "init") {
      // create sub-steps (generateScript, extractScenes, etc.)
    } else if (step.name === "generateScript") {
      // call handleScriptGeneration() ...
    }
    // ...
  };
}

The agent checks the step’s name and dispatches it to the right logic (e.g., generateScript, extractScenes). Each step runs an AI-driven process or data extraction, then updates the step status accordingly.

3. Generating Film Scripts with LangChain

LangChain manages prompts and output parsers. For example, in sceneTechnicalExtractor.ts, we might see:

import { ChatOpenAI } from "@langchain/openai";
import { ChatPromptTemplate } from "@langchain/core/prompts";

export class SceneTechnicalExtractor {
  private scriptChain = /* Sequence definition with ChatOpenAI... */

  async generateScript({ idea, title, lyrics, tags, duration }) {
    // builds a prompt with specialized instructions
    return await this.scriptChain.invoke({ idea, title, lyrics, tags, duration });
  }
}

This approach instructs the AI to produce a screenplay-like text with scenes, durations, camera specs, etc.

4. Validating Steps and Sending Logs

To record events:

import { logMessage } from "../logger/logger";

logMessage(payments, {
  task_id: step.task_id,
  level: "info",
  message: `Processing step: ${step.name}`,
});

This logs both locally and in Nevermined, ensuring a full audit trail for each sub-step.

Development & Testing

1. Development Server

   npm run dev
   

   Copy
   • In some setups, this watches for file changes and restarts automatically.

2. Build

   npm run build
   

   Copy

3. Testing

   npm test
   

   Copy
   • Runs any available unit or integration tests.

License

Apache License 2.0

(C) 2025 Nevermined AG

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at:

   http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions
and limitations under the License.