Building an AI Image Generation Agent with ChatGPT & Playwright

Introduction

Automation is evolving rapidly, and combining browser automation with AI unlocks powerful possibilities. In this article, we’ll walk through a real-world system: an AI Agent that uses ChatGPT to generate prompts and images automatically, while handling confirmations, downloads, and rate limits intelligently.

This is not just a script — it’s a production-style intelligent workflow.

🧠 What This AI Agent Does

• Logs into ChatGPT using a persistent browser session
• Generates high-quality image prompts
• Sends prompts to ChatGPT
• Detects when images are ready
• Downloads images automatically
• Handles confirmation prompts
• Manages rate limits
• Resets context for clean execution

⚙️ System Architecture Overview

User Input → Playwright → ChatGPT → AI Response
                  ↓
          Decision Engine
                  ↓
     Download / Confirm / Retry
                  ↓
           Save + Reset Loop

🔄 Step-by-Step Algorithm

1. Initialization Phase

• Parse CLI arguments (prompt type, count, profile, URL)
• Create required directories (images, logs, prompts)
• Launch Chrome using Playwright persistent session
• Open or restore ChatGPT page

2. Session Handling

• Check if ChatGPT is already logged in
• If not:
  • Wait for manual login
  • Detect active session via chat input

3. Prompt Generation

• If prompts already exist → load them
• Else:
  • Ask ChatGPT to generate prompts in JSON format
  • Extract and validate prompts
  • Save locally for reuse

4. Core Processing Loop

For each prompt:

Step 1: Send Prompt

• Input prompt into ChatGPT
• Wait for response to stabilize

Step 2: Try Direct Download (Highest Priority)

• If download button is available:
  • Download image immediately
  • Save to disk
  • Reset page
  • Move to next prompt

Step 3: Handle Confirmation (Only if Needed)

• If ChatGPT asks:
  • “Would you like me to generate?”
• Then:
  • Send Yes, generate now
  • Wait for response

Step 4: Retry Download

• Try downloading again after confirmation

Step 5: Handle Rate Limits

• Detect messages like:
  • “You’re generating too quickly”
• Extract wait time
• Sleep and retry

Step 6: Recovery

• Reload ChatGPT page
• Retry up to max attempts

5. Reset Strategy (Critical Optimization)

After every successful download:

• Reload ChatGPT page
• Clear previous context
• Start fresh for next prompt

This ensures:

• Better performance
• Lower failure rate
• Reduced UI clutter
• More reliable automation

🧠 Intelligent Decision Logic

🔥 Priority Rule

If download is available → ALWAYS download first. Never confirm or send new prompt.

Confirmation Logic

• Trigger only when explicit phrases detected:
  • “Would you like me to generate?”
  • “Shall I generate?”
• Avoid generic triggers like:
  • “confirm”
  • “?”

Rate Limit Handling

• Detect rate-limit messages
• Extract wait duration dynamically
• Add buffer time
• Resume execution safely

🧩 Key Components

🧠 Mental Model

Ask → Wait → Observe → Decide → Act → Reset → Repeat

🚀 Why This Approach Works

• Mimics human interaction behavior
• Handles real-world UI unpredictability
• Avoids unnecessary actions
• Recovers from failures gracefully
• Maintains clean execution cycles

⚡ Real-World Use Cases

• YouTube thumbnail generation
• AI art pipelines
• Content automation systems
• Social media creatives
• Bulk image generation workflows

🔚 Conclusion

This AI agent is a perfect example of combining:

• AI intelligence (ChatGPT)
• Browser automation (Playwright)
• Smart decision-making logic

The result is a highly efficient, resilient, and scalable automation system.

💡 Final Thought

The real power of AI is not just generating content — it’s building systems that think, decide, and act intelligently.

Let’s build the future 🚀