EDICO | Eliz & Delia's Image Captioner and Organizer

A local, browser-based tool for building character LoRA training datasets with AI-generated captions.

EDICO turns a folder of character images into a clean, captioned training dataset — no command-line juggling, no manual caption typing. A single PowerShell script launches a polished web app where you:

1. Set up a project (vision model, trigger word, character identity)
2. Drop in your images
3. Define a structured caption vocabulary
4. Run an AI captioning pipeline with live progress
5. Review, edit, rearrange, and export in a visual organizer

Everything runs locally on your machine. Your images never leave your computer.

For more documentation, guides and more software go to: gabrielx.com/open-source-mit-ai-tools

---

✨ Features

• Unified web experience — the entire workflow lives in one browser app (no terminal wizards)
• Ollama-powered captioning — works with any vision model (Llava, Qwen-VL, Gemma, etc.)
• Structured captions — consistent, tag-based captions instead of messy free text
• Custom vocabulary — define exactly what the AI describes (pose, expression, framing, lighting...)
• Live pipeline progress — thumbnail grid with per-image status, resume & retry
• Visual organizer — filter by any attribute, multi-select or single-select modes
• Structured image editor — pick terms per category (no raw text editing)
• Rearrange & auto-arrange — drag to order, or auto-sort by priorities before re-sequencing
• Duplicate & resolution badges — spot potential duplicates and non-standard image sizes instantly
• Dark/light theme — follows your system preference
• 100% local & offline — after model download, no internet needed

---

📋 Requirements

Requirement	Notes
Windows	PowerShell 5.1+ (Windows PowerShell or PowerShell 7+)
Ollama	Must be installed and running
A vision model	Pull one first: ollama pull huihui_ai/qwen3-vl-abliterated:8b-instruct (recommended)
Images	.png, .jpg, .jpeg, or .webp

⚠️ No Python, no Node, no pip installs. Just PowerShell + Ollama.

---

🚀 Quick Start

1. Download this repository (or just dataset_tool.ps1 + start.bat)
2. Install Ollama from ollama.com and start it
3. Pull a vision model (see below)
4. Double-click start.bat
5. Your browser opens EDICO — create a project and follow the steps

---

Recommended vision models

# Qwen3 VL (recommended — excellent captioning quality)
ollama pull huihui_ai/qwen3-vl-abliterated:8b-instruct

# Alternative: Qwen2.5 VL
ollama pull huihui_ai/qwen2.5-vl-abliterated:7b

# Alternative: LLaVA
ollama pull llava