🖼️ Inline Image Generation

GPT Researcher supports inline image generation for research reports using Google's Gemini image generation models. This feature creates contextually relevant illustrations that are embedded directly within your research reports.

Overview

When enabled, GPT Researcher will:

1. Analyze research context after gathering information to identify visualization opportunities
2. Pre-generate images before writing the report (for seamless UX)
3. Embed images inline as the report is written - no post-processing delays!

Quick Start

1. Set Environment Variables

# Required: Enable the feature
IMAGE_GENERATION_ENABLED=true

# Required: Your Google API key
GOOGLE_API_KEY=your_google_api_key_here

# Optional: Specify the model (default shown)
IMAGE_GENERATION_MODEL=models/gemini-2.5-flash-image

# Optional: Maximum images per report (default: 3)
IMAGE_GENERATION_MAX_IMAGES=3

# Optional: Image style - "dark" (default), "light", or "auto"
IMAGE_GENERATION_STYLE=dark

2. Run Research

import asyncio
from gpt_researcher import GPTResearcher

async def main():
    researcher = GPTResearcher(
        query="What are the key components of a modern solar panel system?",
        report_type="research_report"
    )
    
    # Images are automatically generated during research
    await researcher.conduct_research()
    
    # Report includes embedded images
    report = await researcher.write_report()
    print(report)

asyncio.run(main())

That's it! Images will be automatically generated and embedded in your report.

How It Works

The Smart Pre-Generation Flow

Research Phase          Image Planning          Report Writing
     │                       │                       │
     ▼                       ▼                       ▼
┌─────────────┐      ┌──────────────┐      ┌─────────────────┐
│ Gather      │      │ LLM analyzes │      │ Report streams  │
│ information │  →   │ context for  │  →   │ with images     │
│ from sources│      │ 2-3 visuals  │      │ already inline! │
└─────────────┘      └──────────────┘      └─────────────────┘
                            │
                            ▼
                     ┌──────────────┐
                     │ Generate all │
                     │ images in    │
                     │ parallel     │
                     └──────────────┘

Key benefits:

• No waiting - Images are generated during research, not after
• Seamless UX - Report streams with images already embedded
• Context-aware - LLM chooses the best visualization opportunities

Configuration Options

Environment Variables

Variable	Default	Description
IMAGE_GENERATION_ENABLED	false	Master switch to enable/disable
GOOGLE_API_KEY	-	Your Google API key (required)
IMAGE_GENERATION_MODEL	models/gemini-2.5-flash-image	Gemini model to use
IMAGE_GENERATION_MAX_IMAGES	3	Maximum images per report
IMAGE_GENERATION_STYLE	dark	Image style: dark, light, auto

Supported Models

Free Tier (Gemini):

Model	Description
models/gemini-2.5-flash-image	Recommended - fast and free
gemini-2.0-flash-exp-image-generation	Experimental variant

Paid Tier (Imagen) - requires Google Cloud billing:

Model	Description
imagen-4.0-generate-001	Highest quality, supports aspect ratios
imagen-4.0-fast-generate-001	Faster generation

Image Styling

Dark Mode (Default)

Images are generated with styling that matches the GPT Researcher UI:

• Dark background (#0d1117)
• Teal/cyan accents (#14b8a6)
• Glowing, futuristic aesthetic
• Professional infographic style

Light Mode

Set IMAGE_GENERATION_STYLE=light for:

• Clean white/light gray backgrounds
• Deep blue and teal accents
• Corporate/professional aesthetic

Auto Mode

Set IMAGE_GENERATION_STYLE=auto for neutral styling that works in any context.

Output

Image Storage

Generated images are saved to:

outputs/images/{research_id}/img_{hash}_{index}.png

Markdown Embedding

Images are embedded using standard markdown syntax:

## System Architecture

![System Architecture Overview](/outputs/images/research_abc123/img_def456_0.png)

The architecture consists of three main components...

Frontend Display

For the Next.js frontend, images are served via the /outputs/ route which proxies to the backend. Images display at 75% width with teal accent borders.

WebSocket Events

When using the web interface, these events are emitted:

Event	Description
image_planning	Analyzing context for visuals
image_concepts_identified	Found N visualization opportunities
image_generating	Generating image X of Y
images_ready	All images generated successfully

Best Practices

1. Enable for detailed reports - Works best with research_report and detailed_report types

2. Monitor API usage - Free tier has daily quotas. Set IMAGE_GENERATION_MAX_IMAGES=2 to conserve

3. Use dark mode - Default styling matches the app and looks professional

4. Review generated images - AI images occasionally need manual review

Troubleshooting

Images Not Generating

1. Verify IMAGE_GENERATION_ENABLED=true
2. Check that GOOGLE_API_KEY is set and valid
3. Ensure model name is correct (include models/ prefix for Gemini)
4. Check logs for API errors

Quota Exceeded

If you see RESOURCE_EXHAUSTED errors:

• Wait until midnight UTC for daily quota reset
• Reduce IMAGE_GENERATION_MAX_IMAGES
• Enable Google Cloud billing for higher quotas
• Create a new Google Cloud project for fresh quota

Images Not Displaying in Frontend

1. Ensure Next.js frontend is configured with the /outputs proxy
2. Check that backend is serving static files from outputs/
3. Verify image paths in the markdown are correct

Disabling Image Generation

To disable completely:

IMAGE_GENERATION_ENABLED=false

Or simply don't set any IMAGE_GENERATION_* variables - the feature is off by default.