Piwigo AI

Beta — This plugin is currently in beta. You may encounter bugs or unexpected behavior.

Transform your Piwigo gallery into an AI-powered smart platform!

Summary

• Prerequisites
• Installation
  • For Users
  • For Developers
• Features
• Dev
• License

Prerequisites

• PHP: 8.4 minimum
• Database: the plugin works on any database Piwigo supports. The vector features (photo embedding, tags indexation, smart and closed tags modes) depend on it:
  • MariaDB 11.7+ (11.8 recommended): every feature.
  • MySQL 9.0+: vectors are stored and tags can be indexed, but the smart and closed tags modes are unavailable (no vector distance function in the Community edition).
  • Older versions: degraded mode, with descriptions, tags (open mode) and OCR only. The overview page tells when the database is not compatible, and can check it again.
• Public galleries (the AI server sends results back): PHP post_max_size of at least 8 MB, for the batches of tag embeddings. Below, they come by polling, more slowly.

Installation

For Users

Coming..

For Developers

1. Clone the Repository:
   • Clone the Piwigo AI repository to your local machine using: git clone https://github.com/Piwigo/piwigo-ai.git
2. Development and Contributions:
   • Make your changes or improvements to the code.
   • See the Dev section for building CSS.
   • Test your changes thoroughly.
   • Feel free to submit a pull request if you wish to contribute your changes back to the project.

Features

• Descriptions: a description of each photo, shown after the regular one if enabled (with an optional prefix).
• Tags, in one of three modes (Configuration > Tags):
  • Open: the AI creates its own tags.
  • Smart: the AI prefers the indexed tags of the gallery, and creates a tag only when it fits the photo better.
  • Closed: the AI only uses the indexed tags of the gallery, it never creates one.
• OCR: the text found in each photo.
• Embedding: a vector of each photo and of its tags, used to choose the tags, and later for similar photos and natural-language search.
• Tags indexation: from the overview, the tags without vector are sent to the AI server in one go; a card follows the progress live. A renamed tag loses its vector until the next indexation.
• Where to analyze: in the uploader (per upload options) and in the batch manager (Piwigo AI action).
• Private or public gallery: a private gallery uploads its photos and polls the results; a public one lets the AI server fetch the photos and send the results back, with polling as a fallback.

Dev

From the plugin directory (plugins/piwigo_ai):

We are experimenting with Tailwind CSS in this plugin as part of an effort to make Piwigo plugin development a more enjoyable experience for developers.

Build the Tailwind CSS for production (commit css/output.css with the templates that need it):

npm run css:build
or
npx @tailwindcss/cli -i css/input.css -o css/output.css --minify

Watch for CSS changes during development:

npm run css:watch
or
npx @tailwindcss/cli -i css/input.css -o css/output.css --minify --watch

Database changes go in numbered migrations (migrations/N-database.php), applied once each when the plugin is installed or updated. Each one checks the schema before changing it, so it can run again after a partial failure.