Build a Tag-Based Search System for Anime-Style Images in C++

Zhirong Yang — Fri, 08 Aug 2025 01:26:33 +0000

Introduction

When you manage a large folder of anime-style illustrations on your PC, you often want to filter images by specific tags (e.g., miko, blonde hair, short hair).
So I built a lightweight web app in C++ that searches a local image folder by tags.

👉 Live demo: http://198.13.48.172:8080/

This post covers:

What the app does and its key features
How to search (with examples)
Tech stack and implementation outline
How tags are generated
Demo and source code links

App Overview

This web tool searches images that already have tag metadata and supports multi-tag queries.

Features:

AND filtering with multiple tags
Negative (exclude) tags like -long_hair
No database required — works on files in a local folder
Backend: C++ | Frontend: simple HTML

Examples

Note: The following images are style-modified versions of the original search results using ChatGPT’s image transformation, to avoid copyright issues.

Miko + Blonde Hair + 1 Girl

miko, 1girl, blonde_hair

Miko but exclude Long Hair and Black Hair

miko, -long_hair, -black_hair

Advanced Search Options

Find images where the blonde_hair score ≥ 0.8

blonde_hair:0.8

Miko + Blonde Hair (≥ 0.9) + exclude Long Hair

miko, blonde_hair:0.9, -long_hair

Exclude images with high explicitness

miko, -explicit:0.3

How Tag Data Is Generated (img2tags + JSON)

Tags are generated automatically with the Python tool img2tags.
It uses a DeepDanbooru-style ViT model and outputs nested JSON containing tags and scores per image.

Example output:

{
    "tags": {
        "9": {
            "general": 0.9189,
            "sensitive": 0.1126,
            "questionable": 0.0131,
            "explicit": 0.0193
        },
        "0": {
            "simple_background": 0.7185,
            "english_text": 0.6080,
            "no_humans": 0.5793,
            "parody": 0.8484,
            "black_background": 0.9437,
            "title_parody": 0.5970,
            "blending": 0.8171,
            "logo_parody": 0.5945
        }
    }
}

Where:

"tags" — root of the structure
"9" — rating-related category (general, explicit, etc.)
"0" — general tags (character, attributes, background, etc.)
Each tag has a score in [0.0, 1.0]

Demo & Source Code

Live demo: http://198.13.48.172:8080/
GitHub repo: https://github.com/Zhirong641/TagSearchWeb

Technical Notes

Backend: simple HTTP server written in C++
No DB: images + tag JSON files on disk are enough
Frontend: HTML + JavaScript (simple search form)
Negative tags supported (NOT filtering)
Results rendered as an image grid

Roadmap

Tag auto-completion & fuzzy search
UX improvements (lazy loading)
Simple UI for tag editing/fixes
Optional SQLite for larger datasets

Closing

If you want to organize your image collection and quickly pull things up by tags, this is handy.
I also hope it’s a useful reference for anyone who wants to try a small web tool in C++.

DEV Community: Zhirong Yang