Build an AI Agent to Auto-Create and Publish YouTube Videos

YouTube is the second largest search engine in the world. But creating videos takes time—scripting, recording, editing, thumbnail design, and publishing. What if an AI agent could handle most of this for you?

In this guide, I’ll show you exactly how to build an AI agent that generates video scripts, creates AI visuals, produces voiceovers, and publishes directly to your YouTube channel.

How It Works: The Video Pipeline

You can build this using Make (Integromat) for no-code, or Python for more control. I’ll cover both.

Step 1: Set Up YouTube API Access

Step 2: Get Other API Keys

Method A: Build with Make (No-Code)

Step 1: Create New Scenario

Step 2: Generate Video Script

Write a YouTube video script about {{topic}}.

Requirements:
- Duration: 8-10 minutes
- Include hook (first 30 seconds to grab attention)
- 4-5 main sections, each 1-2 minutes
- End with call-to-action (subscribe, like, comment)
- Include timestamps for each section
- Add scene descriptions in brackets like [B-roll: city timelapse]
- End with 2-3 suggested video tags

Format:
Title: [Video title]
Hook: [Opening 30 seconds]
Section 1 [0:30-2:00]: [Content and scene]
Section 2 [2:00-4:00]: [Content and scene]
...
Tags: tag1, tag2, tag3

Step 3: Generate Voiceover

Step 4: Generate Video Clips

Step 5: Assemble Video

Step 6: Create Thumbnail

Step 7: Upload to YouTube

Method B: Build with Python

#!/usr/bin/env python3
"""
AI Agent: Auto-Create and Publish YouTube Videos
"""
import requests
import json
import os
import time
from openai import OpenAI
from elevenlabs import client as elevenlabs_client
from google.oauth2.credentials import Credentials
from google_auth_oauthlib.flow import InstalledAppFlow

# ============== CONFIGURATION ==============
OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY", "your-key")
ELEVENLABS_API_KEY = os.environ.get("ELEVENLABS_API_KEY", "your-key")
RUNWAY_API_KEY = os.environ.get("RUNWAY_API_KEY", "your-key")

client = OpenAI(api_key=OPENAI_API_KEY)
elevenlabs = elevenlabs_client(api_key=ELEVENLABS_API_KEY)

# YouTube API scopes
SCOPES = ['https://www.googleapis.com/auth/youtube.upload']

# ============== FUNCTIONS ==============
def generate_script(topic):
    """Generate YouTube video script"""
    prompt = f"""Write a YouTube video script about {topic}.

Requirements:
- 8-10 minute video
- Hook (first 30 seconds)
- 4-5 main sections with timestamps
- Call-to-action at end
- Scene descriptions in brackets [B-roll: description]
- 3-5 video tags

Format your response exactly like this:
TITLE: [Video Title]
TAGS: tag1, tag2, tag3
---
SCRIPT: [Full script text for voiceover]
SCENES: [Scene1]|[Scene2]|[Scene3]

""" response = client.chat.completions.create( model="gpt-4o", messages=[{"role": "user", "content": prompt}] ) return response.choices[0].message.content def generate_voiceover(script_text, output_file="voiceover.mp3"): """Generate AI voiceover""" audio = elevenlabs.generate( text=script_text, voice="Rachel", model="eleven_v2" ) elevenlabs.save(audio, output_file) return output_file def generate_video_clip(description, duration=5): """Generate video clip using Runway""" response = requests.post( "https://api.dev.runwayml.com/v1/gen3_turbo/text_to_video", headers={"Authorization": f"Bearer {RUNWAY_API_KEY}"}, json={ "prompt": description, "duration": duration, "aspect_ratio": "16:9" } ) task_id = response.json()["id"] # Poll for completion (simplified) time.sleep(120) # Wait for generation return f"https://storage.runwayml.com/videos/{task_id}.mp4" def generate_thumbnail(topic): """Generate YouTube thumbnail""" response = client.images.generate( model="dall-e-3", prompt=f"YouTube thumbnail for: {topic}. High contrast, bold text space on left, professional design.", size="1792x1024" ) return response.data[0].url def download_file(url, filename): """Download file from URL""" response = requests.get(url) with open(filename, 'wb') as f: f.write(response.content) return filename def get_youtube_credentials(): """Get YouTube API credentials""" flow = InstalledAppFlow.from_client_secrets_file( 'client_secrets.json', SCOPES) return flow.run_local_server(port=8080) def upload_to_youtube(video_path, title, description, tags, thumbnail_path): """Upload video to YouTube""" # Initialize YouTube API (simplified) youtube = get_youtube_credentials() request_body = { "snippet": { "title": title, "description": description, "tags": tags, "categoryId": "28", # Science & Technology }, "status": { "privacyStatus": "private", "selfDeclaredMadeForKids": False, }, } # Upload video (requires googleapiclient library) # This is simplified - full implementation needs googleapiclient print(f"Would upload {video_path} to YouTube") print(f"Title: {title}") return "video_id_placeholder" # ============== MAIN FUNCTION ============== def main(topic): print(f"Creating video about: {topic}") # 1. Generate script raw = generate_script(topic) title = raw.split("TITLE:")[1].split("TAGS:")[0].strip() tags = raw.split("TAGS:")[1].split("---")[0].strip().split(", ") script = raw.split("SCRIPT:")[1].split("SCENES:")[0].strip() scenes = raw.split("SCENES:")[1].strip().split("|") # 2. Generate voiceover print("Generating voiceover...") voiceover_path = generate_voiceover(script) # 3. Generate video clips print("Generating video clips...") video_clips = [] for i, scene in enumerate(scenes[:5]): # Limit to 5 clips print(f"Generating clip {i+1}/{len(scenes)}...") clip_url = generate_video_clip(scene) clip_path = f"clip_{i}.mp4" download_file(clip_url, clip_path) video_clips.append(clip_path) # 4. Generate thumbnail print("Generating thumbnail...") thumb_url = generate_thumbnail(topic) thumb_path = "thumbnail.png" download_file(thumb_url, thumb_path) # 5. Upload to YouTube print("Uploading to YouTube...") video_id = upload_to_youtube( video_path="final_video.mp4", # Would be assembled video title=title, description=f"Today we explore {topic}.\n\n#Shorts", tags=tags, thumbnail_path=thumb_path ) print(f"Video uploaded! ID: {video_id}") if __name__ == "__main__": main("how AI is changing video creation")

Cost Breakdown

Content Types That Work Best

Common Problems and Fixes

Your Action Checklist

Frequently Asked Questions