🎬 Generate ad image variations using GPT-4, Dumpling AI & Google Drive

⚡ 5,013 views · 🎬 Content Creation & Video

💡 Pro Tip — Since TikTok doesn’t have a public API, getting data usually means fragile HTTP scraping. ScraperNode is a community node that gives you clean JSON for TikTok profiles, videos, and comments — drop it into your workflow and skip the parsing.

Description

🧾 What this workflow does

This workflow takes a reference ad image and brand website, then uses GPT-4, LangChain, and Dumpling AI to generate 10 high-quality image variations for ad testing. These image variations are visually consistent but subtly different in background, mood, lighting, and tone — perfect for performance testing on platforms like Meta Ads or TikTok.

👤 Who is this for

• DTC marketers and brand designers testing ad creatives
• Creative teams automating visual experimentation
• Content agencies using AI for fast ad mockups
• Performance marketers running multivariate testing

⚙️ How to set up

✅ Requirements

You’ll need the following tools set up in n8n:

• Google Drive (OAuth2 credential)
• Google Sheets (OAuth2 credential)
• OpenAI API (for GPT-4 or GPT-4o)
• Dumpling AI API (via HTTP header authentication)

🛠️ Steps to configure

1. Google Sheet Setup

   • Create a sheet with one column: Image URL
   • Update the Sheet ID and tab name in the final Google Sheets node.

2. Drive Setup

   • Create a folder in Google Drive for storing the reference image.
   • Replace the folderId in the “Upload Ad Image to Google Drive” node.

3. Dumpling AI API Key

   • Use n8n’s credential manager (HTTP Header Auth) — do not hardcode the key.

4. OpenAI API Key

   • Required for both image description and LangChain agent prompt generation.

5. Form Inputs Required

   • Brand Name
   • Brand Website
   • Ad Image (upload field)

🧠 How it works

1. A user submits the brand name, website, and a reference ad image through a form.
2. The image is uploaded to Google Drive.
3. GPT-4o describes the image’s visual style (e.g., mood, lighting, composition).
4. GPT-4 analyzes the brand’s website to define its visual aesthetic.
5. A LangChain agent uses both analyses to create 10 tightly scoped variation prompts.
6. Dumpling AI generates a new image for each prompt using its “FLUX.1-pro” model.
7. Each new image’s link is logged into Google Sheets.

🛠️ How to customize

• 🧪 Change prompt logic to experiment with different variations (e.g., theme, season).
• 🎨 Switch image model in Dumpling AI to one that supports your desired style.
• 🔗 Log additional metadata (prompt, timestamp) to Google Sheets.
• 📤 Connect output images to Airtable, Notion, or a review tool like Figma.
• 🎯 Modify GPT system message to reflect a different tone or brand strategy.

This workflow gives creative teams and marketers an instant, AI-powered ad image testing system — built on real brand visuals, not generic stock content.

🔗 Nodes Used

Google Sheets, HTTP Request, Google Drive, AI Agent, OpenAI Chat Model, Structured Output Parser

📥 Import

Download workflow.json and import into n8n: Workflow menu → Import from File

📖 Importing guide · 🔑 Credential setup