TruthGuard AI: Open-Source Multimodal Deepfake Detection and Forensics Platform

TruthGuard AI is an open-source full-stack multimodal forensics platform for deepfake detection, maintained by nasir050298 and released on GitHub on May 26, 2026. The platform integrates React frontend, Express gateway, and FastAPI image detection service, supporting deepfake detection for images, videos, and audio. It includes built-in real Hugging Face models (dima806/deepfake_vs_real_image_detection) and an audit log system. Designed to address deepfake threats posed by generative AI, it provides professional-level multimodal analysis capabilities while balancing immediate usability and scalability.

With the rapid development of generative AI technology, the threshold for producing deepfake content has dropped to an all-time low. Celebrity photos, political speech videos, family voices, etc., can all be tampered with or forged, leading to severe social risks such as disinformation spread, financial fraud, identity theft, and reputation damage. Traditional anti-counterfeiting methods (like visual inspection and simple metadata checks) are no longer sufficient to handle modern AI-generated content. The market urgently needs professional multimodal forensics tools that can analyze images, videos, and audio simultaneously.

TruthGuard AI adopts a microservice architecture, consisting of three layers:

Frontend Layer: Based on React+TypeScript+Vite, it provides functions such as public website, authentication system (JWT login, supporting analyst/reviewer/admin roles), unified dashboard, multimodal forensics interface, report center, case review queue, model lab, and audit logs. The UI uses Tailwind CSS+Framer Motion+Recharts.

Gateway Layer: Based on Node.js+Express, it handles file upload processing, JWT authentication and authorization, route forwarding, report and audit log storage, and Socket.IO real-time notifications.

AI Service Layer: Based on Python+FastAPI, the image detection service runs real Hugging Face models and provides REST API and health check endpoints.

Image Forensics: After users upload images, the Express gateway forwards them to the FastAPI service. Using a pre-trained Vision Transformer model for inference, it returns information such as real/fake probability distribution, highest confidence category, model ID, and inference time.

Video and Audio: Although currently using simulated results, the frontend and backend interfaces are ready, supporting video timeline risk display, audio waveform visualization, etc.

Case Review Workflow: Analyst submits a request → System generates a report → Reviewer examines → Approve/archive. The entire process is recorded in audit logs, reflecting the "human-in-the-loop" design.

Model Lab: Displays the health status of each AI service (real model loading status of image service, fallback status of video/voice services) and supports custom model upload (reserved).

Responsible AI Design: It has a "Responsible AI" page that clarifies model limitations, false positive risks, the necessity of human review, and the distinction between real models and fallback modes, reflecting attention to ethics.

Progressive Deployment Strategy: First validate the end-to-end process of the image module with real models, keep the video/audio module interfaces ready for subsequent iterations, avoid "big bang" releases, and quickly obtain feedback.

Microservice Decoupling: The image detection service runs independently, allowing independent updates and expansion, centralized GPU resource allocation. The gateway focuses on routing and authentication, and different modality detection can be developed and deployed independently.

Local deployment requires running three terminals: 1. Image AI service (Python FastAPI, port 8001); 2. Express backend (Node.js, port 5000); 3. React frontend (Vite, port 5173). Preset demo accounts: Analyst (analyst@truthguard.ai/demo123), Reviewer (reviewer@truthguard.ai/demo123), Admin (admin@truthguard.ai/demo123).

TruthGuard AI demonstrates the complete construction path of an AI security product from 0 to 1, including the full user journey such as public website customer acquisition, dashboard function usage, and report audit compliance. It is not just a technical demo but a product prototype. Reference value for developers: Multimodal architecture design, balance between real models and simulated data, construction of auditable AI systems, and embodiment of responsible AI principles. As deepfake technology evolves, such detection platforms will play an important role in verifying the credibility of digital content.