LLMImageIndexer

LLMImageIndexer is an intelligent image processing and indexing tool that leverages local AI to generate comprehensive metadata for your image collection. This tool uses advanced language models to analyze images and generate captions and keyword metadata.

Screenshot

Features

Intelligent Image Analysis: Utilizes a local AI model to generate a variable number of keywords and a caption for each image.
Metadata Enhancement: Can automatically edit image metadata with generated tags.
Local Processing: All processing is done locally on your machine.
Multi-Format Support: Handles a wide range of image formats, including all major raw camera files.
User-Friendly GUI: Includes a GUI and installer. Relies on Koboldcpp, a single executable, for all AI functionality.
GPU Acceleration: Will use Apple Metal, Nvidia CUDA, or AMD (Vulkan) hardware if available to greatly speed inference.
Cross-Platform: Supports Windows, macOS ARM, and Linux.
Stop and Start Capability: Can stop and start without having to reprocess all the files again.
Keyword Post-Processing: Expand keywords so all synonyms are added to every image with one of the synonyms, or deduplicate keywords by using the most frequently used synonym in place of all matching synonyms.

Installation

Prerequisites

Python 3.8 or higher
KoboldCPP

Windows Installation

Clone the repository or download the ZIP file and extract it.
Install Python for Windows.
Download KoboldCPP.exe and place it in the LlavaImageTagger folder. If it is not named KoboldCPP.exe, rename it to KoboldCPP.exe
Run llmii-run.bat and wait exiftool to install. When it is complete you must start the file again. If you called it from a terminal window you will need to close the windows and reopen it. It will then create a python environment and download the model weights. The download is quite large (6GB) and there is no progress bar, but it only needs to do this once. Once it is done KoboldCPP will start and one of the terminal windows will say Please connect to custom endpoint at http://localhost:5001 and then it is ready.

macOS Installation (including ARM)

Clone the repository or download the ZIP file and extract it.
Install Python 3.7 or higher if not already installed. You can use Homebrew:
```
brew install python
```
Install ExifTool:
```
brew install exiftool
```
Download KoboldCPP for macOS and place it in the LLMImageIndexer folder.
Open a terminal in the LLMImageIndexer folder and run:
```
chmod +x koboldcpp-mac-arm64
./llmii-run.sh
```

Linux Installation

Clone the repository or download and extract the ZIP file.
Install Python 3.7 or higher if not already installed. Use your distribution's package manager, for example on Ubuntu:
```
sudo apt-get update
sudo apt-get install python3 python3-pip
```

Install ExifTool. On Ubuntu:

sudo apt-get install libimage-exiftool-perl

Download the appropriate KoboldCPP binary for your Linux distribution from KoboldCPP releases and place it in the LLMImageIndexer folder.
Open a terminal in the LLMImageIndexer folder and run:
```
chmod +x koboldcpp-linux-x64
./llmii-run.sh
```

For all platforms, the script will set up the Python environment, install dependencies, and download necessary model weights (6GB total). This initial setup is performed only once and will take a few minutes depending on your download speed.

Usage

Launch the LLMImageIndexer GUI:
- On Windows: Run llmii-run.bat
- On macOS/Linux: Run python3 llmii-gui.py
Ensure KoboldCPP is running. Wait until you see the following message in the KoboldCPP window:
```
Please connect to custom endpoint at http://localhost:5001
```
Configure the indexing settings in the GUI:
- Select the target image directory
- Set the API URL (default: http://localhost:5001)
- Choose metadata tags to generate (keywords, descriptions)
- Set additional options (crawl subdirectories, backup files, etc.)
Click "Run Image Indexer" to start the process.
Monitor the progress in the output area of the GUI.

Configuration Options

Directory: Target image directory (includes subdirectories by default)
API URL: KoboldCPP API endpoint (change if running on another machine)
API Password: Set if required by your KoboldCPP setup
Caption: Have the LLM describe the image and set it in XMP:Description (doubles processing time)
GenTokens: Amount of tokens for the LLM to generate
Skip processed files not in database: Will not attempt to reprocess files with a UUID and keywords even if they are not in the llmii.json database
Reprocess failed: If any files failed in the last round, it will try to process them again
Reprocess ALL: The files that are processed already are stored in a database and skipped if you resume later, this will do them all over again
Don't crawl subdirectories: Disable scanning of subdirectories
Don't make backups before writing: Skip creating backup files (NOTE: this applies to processing and post-processing; if you enable post processing and leave this unchecked it will make a second backup!)
Pretend mode: Simulate processing without writing to files or database
Skip processing: If you want to use the keyword processing and don't want it to check every image in the directory before starting, check this box
Keywords: Choose to clear and write new keywords or update existing ones
Post processing keywords: Keep keywords as generated, Expand keywords by applying all synonyms to matching keywords, or Dedepe keywords by replacing matching synonyms with most frequent synonym; these option take place after the indexer has completed unless the 'skip processing' box is checked

More Information and Troubleshooting

Consult the wiki for more information and troubleshooting steps.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgements

ExifTool for metadata manipulation
KoboldCPP for local AI processing
PyQt6 for the GUI framework
Fix Busted JSON and Json Repair for help with mangled JSON parsing

jabberjabberjabber / LLavaImageTagger

readme