AI Image Tagger
Eagle plugin for automatic image tagging using local ONNX models, CLIP, and LLMs
Overview
AI Image Tagger is a plugin for Eagle (a media library app) that automates image tagging using a multi-model pipeline. It supports fully offline inference via WD14 ONNX models, semantic CLIP-based suggestions matched against the user's existing tag library, and cloud or local LLM tagging via OpenAI, Anthropic, or any OpenAI-compatible server like Ollama. Built out of a personal need to manage large reference libraries without manual tagging.
Motivation
Eagle is a tool I've used extensively for storing reference images and other media — the browser plugin makes saving content effortless. But tagging is a different story. For a new library without any tags, or a large one with thousands of untagged images, manually organizing everything is tedious. I built this plugin to solve that for myself: let AI handle the tagging, whether through fast offline models or a full LLM, so the library stays organized without the manual overhead. WD14 was the first model I integrated because it matched my personal use case well (Danbooru-style tags for reference art), but LLM support was added later to make it useful for a wider range of image types.
Highlights
- ▸Integrated WD14 ONNX model inference for fast, fully offline Danbooru-style image tagging with configurable general/character thresholds and top-N controls.
- ▸Built a CLIP-based suggestion engine that ranks the user's existing Eagle tag library by semantic similarity to each image, surfacing relevant tags without retraining.
- ▸Implemented LLM tagging with support for OpenAI, Anthropic, and local OpenAI-compatible servers (LM Studio, Ollama), using vision models to generate structured tag output.
- ▸Developed a gallery view with multi-select, batch tagging with progress tracking, per-tag filtering (AND logic), and cancellation — handling large libraries without blocking the UI.
- ▸Added prompt presets, a global tag blacklist, and auto-save to give power users fine-grained control over inference behavior across different image types.
- ▸Built tag deduplication to prevent re-adding tags already saved in Eagle, keeping libraries clean during batch operations.
Screenshots
