e621tagger

e621tagger is a web-based tool that automatically generates relevant tags for furry artwork using the JTP-3 Hydra model by Project RedRocket.

Can be accessed here: tagger.fenrir784.ru

For usage tips click Need Help? in settings.

Features:

• Mobile and desktop interface
• e621\PostyBirb tag formats
• PWA support
• Adjustable thresholds
• e621wiki lookups
• Self-host-able with all-in-one docker image

Have problems or suggestions? Contact me directly via Telegram

⚠️ Note: This repository contains the source code for the app. A live, fully functional instance is available at the link above.

---

🐳 Run with Docker (Self‑Hosted)

Example docker-compose.yml

services:
  e621tagger:
    image: ghcr.io/fenrir784/e621tagger:latest
    container_name: e621tagger
    ports:
      - "5000:5000"
    environment:
      - TZ=Europe/Moscow           # optional, adjust to your timezone
      - SAVE_UPLOADS=true          # optional, default is false
      - USE_PROXY=true             # optional, default is false, set to true for correct client IPs while using reverse proxy
      - GUNICORN_WORKERS=2         # optional, default is 1, amount of workers up, useful for redundancy  
      - GUNICORN_TIMEOUT=120       # optional, default is 120(seconds), timeout before worker tries to restart
    volumes:
      - ./uploads:/app/uploads     # where uploaded images are saved (if enabled)
    restart: unless-stopped

1. Save the file as docker-compose.yml.
2. Run docker-compose up -d.
3. Access the app at http://localhost:5000.

---

🧠 How It Works

• The backend uses PyTorch and the JTP-3 Hydra model, a fine‑tuned image classifier trained on e621 data.
• When you upload an image, it is resized, converted to patches, and fed through the model.
• The model outputs confidence scores for over 7,500 possible tags; the adjustable amount of tags are returned.
• All processing happens on the server; no data is sent to third parties.

---

🛠️ Technology Stack

• Backend: Python, Flask, PyTorch, Gunicorn
• Frontend: Flask rendered HTML, CSS, JavaScript
• Model: RedRocket/JTP-3 on Hugging Face

---

📄 License

This project is licensed under the Apache-2.0 license. See the LICENSE file for details.

---

Created by fenrir784.ru – if you like the tool, consider dropping a star ⭐ on GitHub!