Flash Insight 📸

Real-time visual AI processing at your fingertips. Flash Insight instantly analyzes screen content using Google's Gemini Vision AI, providing immediate, accurate insights from any visual information.

Overview

Flash Insight combines advanced computer vision with real-time screen capture to transform visual information into instant understanding. Built with PyQt5 and Gemini Vision AI, it offers a clean, efficient interface for rapid visual analysis.

Flash Insight shown alongside an iPhone screen mirror, demonstrating real-time quiz analysis and answer processing.

Features

• 📸 Real-time preview of capture area
• 🎯 Adjustable capture region
• 🤖 AI-powered content analysis using Google's Gemini
• 🖥️ Clean, modern interface
• ⚡ Fast and responsive processing
• 🔝 Always-on-top window for easy access

🚀 Getting Started

Prerequisites

• Python 3.8+
• Google Gemini API key (free tier available)

Installation

1. Clone the repository

git clone https://github.com/JadAssaf/flash-insight.git
cd flash-insight

2. Install dependencies

pip install -r requirements.txt

3. Create a .env file and add your Google API key:

echo "GOOGLE_API_KEY=your_api_key_here" > .env

Note: You can get your Google API key from here.

Usage

Launch the application:

python flash-insight.py

The interface provides precise controls for capture area adjustment and real-time preview of the analyzed region.

Configuration

Configure the model and generation parameters in config.py

Model Selection

MODEL_NAME = 'gemini-2.0-flash'  # Default: Fast responses

Available models:

• gemini-2.0-flash: Standard fast model for quick responses
• gemini-2.0-flash-lite: Lighter version of Flash, potentially faster but may be less capable
• gemini-2.0-pro-exp-02-05: Experimental Pro version with potential improvements
• gemini-2.0-flash-thinking-exp-01-21: Experimental Flash model with enhanced thinking capabilities

Check Google AI Studio for the most up-to-date model options, as available models may change over time.

Generation Settings

The behavior of the AI can be customized through these settings in GENERATION_CONFIG:

GENERATION_CONFIG = {
    "temperature": 0.1,        # Controls response randomness (0.0 - 1.0)
    "candidate_count": 1,      # Number of responses to generate
    "max_output_tokens": 20,   # Maximum response length
}

Temperature (0.0 - 1.0)

• Low (0.0 - 0.3): More precise, consistent answers
  • Best for: Quiz answers, factual responses
  • Default: 0.1
• Medium (0.4 - 0.6): Balanced between consistency and creativity
  • Best for: General purpose use
• High (0.7 - 1.0): More creative, varied responses
  • Best for: Creative writing, brainstorming

Candidate Count

• Controls how many different answers the model generates
• Higher values = more options but increased API usage
• Default: 1 (single best answer)

Max Output Tokens

• Controls response length
• 20 tokens ≈ 15-20 words
• Increase for longer explanations
• Decrease for more concise answers

Example Configurations

1. For Detailed Explanations

GENERATION_CONFIG = {
    "temperature": 0.3,
    "candidate_count": 1,
    "max_output_tokens": 50,
}

2. For Creative Responses

GENERATION_CONFIG = {
    "temperature": 0.7,
    "candidate_count": 1,
    "max_output_tokens": 30,
}

3. For Multiple Answer Suggestions

GENERATION_CONFIG = {
    "temperature": 0.5,
    "candidate_count": 3,
    "max_output_tokens": 20,
}

Prompt Engineering

The system prompt in config.py can be modified to alter the AI's interpretation and response patterns. Consider:

• Response formatting
• Analysis parameters
• Question type handling
• Context specifications

License

MIT License - feel free to use and modify as needed!