Automating YouTube Content with n8n and LLMs

Managing the full YouTube content lifecycle — from idea to publishing — is repetitive and time-consuming. To address this, I built an automated pipeline that generates scripts, creates audio and visuals, merges audio and video with FFmpeg, and uploads the final video to YouTube. The workflow integrates custom voice generation, image creation, a custom merging tool, and automated upload.

This workflow was developed using n8n, a powerful no/low-code automation tool, together with external APIs such as OpenAI, ElevenLabs, and Runway ML. Below, I’ll walk you through the pipeline step by step.

Tech Stack Used

• OpenAI API → Generates script, title, and description

• ElevenLabs API → Creates the voice-over

• Runway ML API → Generates images and videos

• Custom API (Node.js + FFmpeg) → Merges audio and video

• YouTube API → Uploads the final video

Step 1: User Input

The workflow begins with an n8n Form Trigger, where the user describes the idea and sets requirements for the video.
For example: “What is Agentic AI? Duration: 10 seconds.”

Prompt Example: Describe the process of Robotic Process Automation. Duration should be 10 seconds. Upload it to YouTube.

Step 2: Script Generation

The user’s input is passed to OpenAI, which generates a complete video script based on the provided idea. The input includes the concept, desired duration, and target platform. The OpenAI API then returns a title, script, and other metadata.

API Key Setup:

Instead of hardcoding API keys in nodes, place them in a .env file so the workflow can securely access them.

1. 1. Create a .env file and store all keys there.

   2. In the OpenAI API Node, go to Credentials, set the Key Name as Authorization, and its value as {{$env.YOUR_KEY_NAME_IN_.ENV}}.

   3. Ensure the key name matches exactly between the expression and the .env file.

Step 3: Data Structuring

A Code Node parses the JSON response returned by OpenAI:

let content = items[0].json.message.content;

if (typeof content === 'string') {
// 1. Remove Markdown code fences
content = content.replace(/(?:json)?\s*([\s\S]*?)\s*/i, '$1');

// 2. Remove leading/trailing whitespace and \n, \r
content = content.trim().replace(/^\\n|\\n$/g, '');

try {
// 3. Parse the JSON
content = JSON.parse(content);
} catch (err) {
return [{
json: {
raw: items[0].json.message.content,
cleaned: content,
error: 'Failed to parse JSON: ' + err.message
}
}];
}
}

This node (“Code Node”) parses the JSON returned by the OpenAI API Node.
Next, a Set Node extracts key values such as script text, title, video duration, and target platform.

At this stage, the workflow branches into two paths: one for audio generation and another for visual generation.