Title: DetectZoo: A Comprehensive Framework for Detecting AI-Generated Content in Text, Audio, and Images

Abstract

As generative models become increasingly sophisticated and widespread, the line separating human-created material from machine-generated output is blurring. This shift has spurred significant research into detection methods spanning text, visual, and audio formats. However, current solutions often present substantial hurdles: commercial tools are typically closed, while open-source alternatives frequently suffer from fragmented codebases, unique preprocessing requirements, and inconsistent evaluation standards. These disparities complicate adoption, hinder fair comparisons, and make replication of results challenging.

To bridge this critical gap, we present DetectZoo, an innovative and extensible toolkit that establishes a unified interface for detecting AI-generated content across multiple modalities. DetectZoo streamlines the entire empirical workflow—ranging from data ingestion and preprocessing to model assessment—providing researchers with a cohesive platform to systematically benchmark leading detection algorithms. By consolidating various public datasets and baseline detection models under a single, standardized API, the toolkit ensures rigorous and reproducible evaluation processes.

Key features of DetectZoo include reference implementations for 61 detectors and native loaders for 22 benchmark datasets. It employs a standardized evaluation pipeline that delivers multiple metrics through a consistent interface. Each detector operates as a self-contained unit, accessible via a uniform interface, and automatically caches pretrained weights to ensure accurate reproduction of originally published results. By lowering the entry barrier for multi-modal AI forensics, DetectZoo empowers researchers to pinpoint performance disparities across different domains and accelerates the creation of robust, generalizable detection methods. The project’s open-source code and detailed documentation are available at https://github.com/sadjadeb/DetectZoo, and the package can be easily installed using the command pip install detectzoo.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC