LFM2-VL Image Analyzer

Download:

• LFM2-VL Image Analyzer v0.1.0 (Windows) - ~86 MB

---

⚠️ EARLY PROTOTYPE - A Windows desktop application for analyzing images using LiquidAI's LFM2-VL vision models.

⚠️ Important Notice

This is an early prototype software. Please be aware:

• May not work on all systems - Particularly older hardware or certain Windows configurations
• CPU Compatibility Issues - Some older processors (Intel i7-4790 and similar) may experience crashes
• Windows 10 vs 11 - Better stability on Windows 11, mixed results on Windows 10
• Resource Intensive - Requires significant RAM and CPU resources
• Limited Support - This is experimental software without warranty

System Requirements

Minimum Requirements

• Windows 10/11 (64-bit) - Windows 11 preferred
• 8GB RAM minimum (16GB+ recommended)
• Modern CPU - Intel Core i5-8xxx+ or AMD Ryzen 5 2xxx+ recommended
• 4GB free disk space

Known Problematic Hardware

• Intel i7-4790 (Haswell 2014) and older processors
• Systems with less than 8GB RAM
• Very old Windows 10 installations

Features

• Image analysis using LiquidAI LFM2-VL vision-language models
• Support for LFM2-VL-450M (recommended) and LFM2-VL-1.6B models
• Multiple quantization options (Q4_0, Q8_0)
• Modern GUI built with PySide6
• Automatic CPU detection and optimization
• Customizable prompts and analysis parameters

Quick Start

1. Download the latest LFM2-VL-Analyzer.exe from releases
2. Download Models from HuggingFace (see Model Setup below)
3. Run the executable - it will create necessary directories
4. Configure model paths in the application
5. Test with a simple image first

Model Setup

You need to download GGUF models from HuggingFace:

LFM2-VL-450M (Recommended)

• Main model: LiquidAI/LFM2-VL-450M-GGUF
• Required files:
  • LFM2-VL-450M-Q4_0.gguf (219 MB) - fastest
  • mmproj-LFM2-VL-450M-Q8_0.gguf - vision projector
• Memory: ~1GB RAM required

LFM2-VL-1.6B (Advanced)

• Main model: LiquidAI/LFM2-VL-1.6B-GGUF
• Required files:
  • LFM2-VL-1.6B-Q4_0.gguf (696 MB)
  • mmproj-LFM2-VL-1.6B-Q8_0.gguf - vision projector
• Memory: ~2.5GB RAM required

Model Placement

Place downloaded .gguf files in:

• Bundled version: %APPDATA%\LFM2-VL-Analyzer\models\

Architecture

This application uses an innovative HTTP API approach:

• Persistent llama-server process for stability
• HTTP requests instead of subprocess calls
• Eliminates Windows process creation stress
• Automatic CPU optimization and compatibility detection

Troubleshooting

Application Won't Start

• Ensure Windows Defender isn't blocking the executable
• Try running as Administrator
• Check Windows Event Viewer for crash details

Server Crashes on Older CPUs

• The application automatically detects CPU capabilities
• Older processors get conservative "compatibility mode"
• If crashes persist, your CPU may be incompatible

Out of Memory Errors

• Try the smaller LFM2-VL-450M model
• Close other applications to free RAM
• Reduce batch sizes in configuration

General Issues

1. Check logs in %APPDATA%\LFM2-VL-Analyzer\logs\
2. Try the LFM2-VL-450M model first (lighter)
3. Ensure all required model files are downloaded
4. Restart the application if analysis gets stuck

Configuration

Settings stored in:

• Production: %APPDATA%\LFM2-VL-Analyzer\config.json
• Development: ./lfm2_vl_analyzer/config.json

License

Freeware - Free for personal use. See LICENSE file for full terms.

• Copyright: Filip Bogacz (Fibogacci) 2025
• Commercial use: Prohibited without permission
• Distribution: GitHub releases only
• Model licensing: Users responsible for HuggingFace model compliance

Development Status

• Version: v0.1.0 (Early Prototype)
• Status: Experimental, limited compatibility
• Target: Windows standalone distribution
• Feedback: Welcome via GitHub issues

Contributing

This is experimental software. Bug reports and compatibility feedback are valuable:

1. Include your Windows version and CPU model
2. Attach log files from %APPDATA%\LFM2-VL-Analyzer\logs\
3. Specify which model you were using

Contact

Developer: Filip Bogacz (Fibogacci)

• Website: https://fibogacci.com
• GitHub: https://github.com/Fibogacci
• LinkedIn: https://linkedin.com/in/Fibogacci
• Medium: https://medium.com/@Fibogacci
• Buy Me a Coffee: https://buymeacoffee.com/Fibogacci

---

⚠️ Disclaimer: This software is provided "as-is" without warranty. Use at your own risk.