Generate social videos with AI avatars using VEED, Claude, and OpenAI

Generate social videos with AI avatars using VEED and Claude

Overview

This n8n workflow automatically generates TikTok/Reels-ready talking head videos from scratch. You provide a topic and intention, and the workflow handles everything: scriptwriting, avatar generation, voiceover creation, and video rendering.

Output: Vertical (9:16) AI-generated videos with lip-synced avatars, ready for social media posting.

---

What It Does

Topic + Intention → Claude writes script → OpenAI generates avatar → OpenAI creates voiceover → VEED renders video → Saved to Google Drive + logged to Sheets

Pipeline Breakdown

Step	Tool	What Happens
1. Script Generation	Claude Sonnet 4	Creates hook, script (30-45 sec), caption, and image prompt based on your topic and intention
2. Avatar Generation	OpenAI gpt-image-1	Generates photorealistic portrait image (1024×1536)
3. Voiceover	OpenAI TTS-1-HD	Converts script to natural speech (Nova voice)
4. Video Rendering	VEED Fabric 1.0	Lip-syncs avatar to audio, creates final video
5. Storage	Google Drive	Uploads final MP4
6. Logging	Google Sheets	Records all metadata (script, caption, URLs, timestamps)

---

Required Connections

API Keys (entered in Configuration node)

Service	Key Type	Where to Get
Anthropic	API Key	https://console.anthropic.com/settings/keys
OpenAI	API Key	https://platform.openai.com/api-keys

> ⚠️ OpenAI Note: gpt-image-1 requires organization verification. Go to https://platform.openai.com/settings/organization/general to verify.

n8n Credentials (connect in n8n)

Node	Credential Type	Purpose
** Generate Video (VEED)**	FAL.ai API	VEED video rendering
** Upload to Drive**	Google Drive OAuth2	Store final videos
** Log to Sheets**	Google Sheets OAuth2	Track all generated content

---

Configuration Options

Edit the ⚙️ Workflow Configuration node to customize. The configuration uses a JSON format:

{
 "topic": "AI video creation tools",
 "intention": "informative",
 "brand_name": "YOUR_BRAND_NAME",
 "target_audience": "content creators and marketers",
 "trending_hashtags": "#AIvideo #ContentCreation #VideoMarketing #AItools #TikTokTips",
 "num_videos": 1,
 "anthropic_api_key": "YOUR_ANTHROPIC_API_KEY",
 "openai_api_key": "YOUR_OPENAI_API_KEY",
 "video_resolution": "720p",
 "video_aspect_ratio": "9:16",
 "custom_avatar_description": "",
 "custom_script": ""
}

Configuration Fields Explained

Field	Required	Description
topic	✅	The subject of your video (e.g., "AI productivity tools")
intention	✅	Content style: informative, lead_generation, or disruption
brand_name	✅	Your brand/product name to mention
target_audience	✅	Who you're creating content for
trending_hashtags	✅	Hashtags to include in the caption
num_videos	✅	How many videos to generate (1-5 recommended)
anthropic_api_key	✅	Your Anthropic API key
openai_api_key	✅	Your OpenAI API key
video_resolution	✅	Video quality: 720p or 1080p
video_aspect_ratio	✅	Aspect ratio: 9:16 (vertical) or 16:9 (horizontal)
custom_avatar_description	❌	Optional: Describe your avatar (leave empty for AI-generated)
custom_script	❌	Optional: Your own script (leave empty for AI-generated)

Intention Types

Intention	Content Style	Best For
informative	Educational, value-driven, builds trust	Thought leadership, tutorials
lead_generation	Creates curiosity, soft CTA	Product awareness, funnels
disruption	Bold, provocative, scroll-stopping	Viral potential, brand awareness

---

Custom Avatar & Script Options

The workflow supports flexible content generation - you can let Claude generate everything, or provide your own inputs.

Custom Avatar Description

Leave custom_avatar_description empty to let Claude decide, or provide your own:

"custom_avatar_description": "a female influencer in her 30s, with a coworking space in the background, attractive but charismatic"

Examples:

• "a woman in her 20s with gym clothes"
• "a bearded man in his 30s wearing a hoodie"
• "a professional woman with glasses in business casual"

Custom Script

Leave custom_script empty to let Claude write it, or provide your own:

"custom_script": "This is my custom script. VEED is a great platform for creating videos like this. You can try it too!"

Guidelines for custom scripts:

• Keep it 30-45 seconds when read aloud
• Maximum ~450 characters
• Avoid special characters for TTS compatibility
• Write naturally, as if speaking

Behavior Matrix

custom_avatar_description	custom_script	What Claude Generates
Empty	Empty	Avatar + Script + Caption
Provided	Empty	Script + Caption
Empty	Provided	Avatar + Caption
Provided	Provided	Caption only

---

Content Angles (auto-rotated)

When generating multiple videos, the workflow automatically varies the approach:

#	Angle	Hook Style
1	Problem-solution	Opens with a question
2	Myth-busting	Opens with controversial statement
3	Quick-tip	Opens with a number/statistic
4	Before-after	Opens with transformation
5	Trend-commentary	Opens with news/timely angle

---

Output

Per Video Generated

Asset	Format	Location
Final Video	MP4 (720p, 9:16)	Google Drive folder
Avatar Image	PNG (1024×1536)	tmpfiles.org (temporary)
Voiceover	MP3	tmpfiles.org (temporary)
Metadata	Row entry	Google Sheets

Google Sheets Columns

Column	Description
topic	Video topic
intention	Content intention used
brand_name	Brand mentioned
content_theme	2-3 word theme summary
script_audio	Full voiceover script
script_image	Image generation prompt
caption	Ready-to-post TikTok caption with hashtags
image_url	Temporary avatar image URL
audio_url	Temporary audio URL
video_url	Google Drive link to final video
status	done/error
created_at	Timestamp

---

Estimated Costs Per Video

Service	Usage	Approximate Cost
Claude Sonnet 4	~2K tokens	~$0.01
OpenAI gpt-image-1	1 image (1024×1536)	~$0.04-0.08
OpenAI TTS-1-HD	~450 characters	~$0.01
VEED/FAL.ai	1 video render	~$0.10-0.20
Total		~$0.15-0.30 per video

> Costs vary based on script length and current API pricing.

---

Setup Checklist

Step 1: Import Workflow

• Import generate-social-videos-with-ai-avatars-using-veed-and-claude.json into n8n

Step 2: Configure API Keys

• Open the ⚙️ Workflow Configuration node
• Replace YOUR_ANTHROPIC_API_KEY with your actual Anthropic API key
• Replace YOUR_OPENAI_API_KEY with your actual OpenAI API key
• Verify your OpenAI organization at https://platform.openai.com/settings/organization/general (required for gpt-image-1)

Step 3: Connect n8n Credentials

• Click on ** Generate Video (VEED)** node → Add FAL.ai credential
• Click on ** Upload to Drive** node → Add Google Drive OAuth2 credential
• Click on ** Log to Sheets** node → Add Google Sheets OAuth2 credential

Step 4: Configure Storage

• Update the ** Upload to Drive** node with your Google Drive folder URL
• Update the ** Log to Sheets** node with your Google Sheets URL
• Create column headers in your Google Sheet: topic, intention, brand_name, content_theme, script_audio, script_image, caption, image_url, audio_url, video_url, status, created_at

Step 5: Customize Content

• Update topic, brand_name, target_audience, and trending_hashtags
• Optionally add custom_avatar_description and/or custom_script

Step 6: Test

• Set num_videos: 1 for initial testing
• Execute the workflow
• Check Google Drive for the output video
• Verify metadata was logged to Google Sheets

---

MCP Integration (Optional)

This workflow can also be exposed to Claude Desktop via n8n's Model Context Protocol (MCP) integration, allowing you to generate videos through natural language prompts.

To enable MCP:

1. Add a Webhook Trigger node to the workflow (in addition to the Manual Trigger)
2. Connect it to the same ⚙️ Workflow Configuration node
3. Go to Settings → Instance-level MCP → Enable the workflow
4. Configure Claude Desktop with your n8n MCP server URL

Claude Desktop Configuration (Windows):

{
 "mcpServers": {
 "n8n-mcp": {
 "command": "supergateway",
 "args": [
 "--streamableHttp",
 "https://YOUR_N8N_INSTANCE.app.n8n.cloud/mcp-server/http",
 "--header",
 "authorization:Bearer YOUR_MCP_ACCESS_TOKEN"
 ]
 }
 }
}

> Note: Install supergateway globally first: npm install -g supergateway

---

Limitations & Notes

Technical Limitations

• tmpfiles.org: Temporary file URLs expire after ~1 hour. Final videos are safe in Google Drive.
• VEED processing: Takes 2-5 minutes per video depending on length.
• n8n Cloud network: Some external domains are blocked; workflow uses base64 for images to avoid this.

Content Considerations

• Scripts are optimized for 30-45 seconds (TTS-friendly)
• Avatar images are AI-generated (not real people)
• Captions include hashtags automatically
• Each video in a batch gets a different content angle

Best Practices

1. Start small: Test with 1 video before scaling to 5
2. Review scripts: Claude generates good content but review before posting
3. Monitor costs: Check API usage dashboards weekly
4. Backup sheets: The Google Sheet serves as your content database

---

Troubleshooting

Issue	Solution
"Organization must be verified"	Verify at platform.openai.com/settings/organization/general
VEED authentication error	Re-add FAL.ai credential to VEED node
Google Drive "no binary field"	Ensure Download Video outputs to field named data
JSON parse error from Claude	Workflow has fallback content; check Claude node output
Image URL blocked	Workflow uses base64 to avoid this; ensure gpt-image-1 model
MCP "Server disconnected" (Windows)	Install supergateway globally: npm install -g supergateway
MCP path error on Windows	Use supergateway directly instead of npx

---

Version History

Version	Date	Changes
2.0	Jan 2026	Added custom avatar/script options, MCP integration support, improved configuration
1.0	Jan 2026	Initial release with portrait mode, gpt-image-1, native VEED node

---

Credits

Built with:

• n8n - Workflow automation
• Anthropic Claude - Script generation
• OpenAI - Image & audio generation
• VEED Fabric - AI video rendering
• Google Workspace - Storage & logging