ImageTagger

ImageTagger is a desktop annotation tool for image/text pairs, built with PyQt6 for ML dataset curation workflows.

It is designed for teams and solo practitioners who need to keep large caption/tag corpora clean, consistent, and model-ready.

Quick Start

Windows:

1. Double-click install.bat (or run it from a terminal).
2. Double-click run.bat to launch.

Linux / macOS:

1. chmod +x install.sh run.sh update.sh
2. ./install.sh
3. ./run.sh

To update dependencies later: run update.bat (Windows) or ./update.sh (Linux/macOS).

Then:

1. Open a folder containing images.
2. Start an LLM server endpoint:
   • Ollama (default): https://ollama.com on port 11434.
   • OpenAI-compatible server (for example vLLM) on a non-11434 port such as 8000.
3. Enter the endpoint and fetch models.
4. Qwen3-VL-8B is the current recommended model for the best practical performance/quality balance.
5. Select one or more images.
6. Generate, validate, and resolve fixups.

Supported image formats: jpg, jpeg, png, bmp, gif, webp.

Documentation

• Docs index: docs/README.md
• Full usage guide: docs/usage.md
• Ollama settings: docs/ollama_settings.md
• Screenshots: docs/screenshots/main_window.png, docs/screenshots/merge_dialog.png

Highlights

• Batch operations on selected images for Generate, Validate, and AI Find.
• Fixup workflow backed by per-image .fixup files.
• Expression-based filtering with fixup, quoted tags, freetext, AND/OR, and parentheses.
• Editable prompt tabs for Description, Tagging, Validation, and AI Search.
• Preview context menu with Open in Default App and Open With support.

Screenshots

Main window:

Merge dialog:

Acknowledgement and Inspiration

This project is heavily inspired by TagGUI.

TagGUI deserves full credit for the core UI layout direction and practical workflow ideas.

AI Generation Disclosure

For transparency and legal disclosure, this codebase was 100% AI-generated with GitHub Copilot.