Parallax is a state-of-the-art, end-to-end ecosystem designed to combat digital misinformation by providing forensic-grade analysis of content across all major media formats. Powered by a suite of custom deep learning models, this project delivers a robust, scalable, and transparent solution for verifying the authenticity of the digital world.
This repository is the central hub for all components of the Parallax project, from the backend APIs and advanced AI models to the user-facing frontend and browser extension.
Parallax is architected as a modular, multi-modal detection platform where each modality operates independently while contributing to a unified analysis pipeline. This design allows continuous model improvement without disrupting other system components.
We integrate advanced language modeling and claim verification techniques to create an authoritative verification engine.
- A language model identifies the central claim within submitted text.
- The claim is evaluated through structured verification and evidence retrieval pipelines.
- Results are synthesized into a clear, evidence-backed explanation.
Rather than labeling content as simply “fake” or “real,” Parallax provides contextual reasoning and verified insights to improve transparency and trust.
Parallax analyzes media across video, image, audio, and text formats using specialized AI pipelines.
A dual-layer forensic framework:
- Behavioral Analysis – Detects unnatural facial movements and micro-expressions.
- Biological Signal Analysis (rPPG) – Extracts physiological pulse signals from facial regions to detect synthetic inconsistencies.
- Specialized detection model for human faces.
- Zero-shot generalized model for non-face AI-generated images.
- Identifies GAN-based and diffusion-based synthetic artifacts.
- Lightweight model using MFCC-based feature extraction.
- Detects synthetic speech by analyzing inconsistencies in voiceprints and acoustic signatures.
- Hybrid stylistic + semantic classifier.
- Evaluates linguistic entropy, coherence patterns, and structural artifacts associated with AI-generated text.
- Integrated into the backend API for seamless processing.
Each directory contains a dedicated README.md with detailed technical information and usage instructions for that module.
| Component | Directory Path | Description & Key Technologies |
|---|---|---|
| Video Modality | ./video_modality_raw_files/ |
Deepfake detection using behavioral and biological (rPPG) signal analysis. |
| Image Modality | ./Image_modality_raw_files/Zero Shot Detection/ |
Face-specific and generalized zero-shot synthetic image detection. |
| Audio Modality | ./audio_modality_raw_files/Deepfake Audio Detection/ |
MFCC-based synthetic voice detection model. |
| Text Misinformation | ./text_modality(misinformation)/ |
Hybrid semantic + stylistic AI-generated text detection system. |
| Backend | ./Backend/ |
Central API handling model orchestration and response aggregation. |
| Web Extension | ./Web-Extension/ |
Browser extension enabling real-time misinformation detection while browsing. |
- Clone the repository:
git clone https://github.com/your-username/Parallax.git cd Parallax - Navigate to the
Backend/directory and follow its setup guide to deploy the core API. - Proceed to the
Frontend/and Web-Extension/ directories to set up the user-facing components.