About
Deepfake-O-Meter (Deep-O-Meter) is a web-based deepfake detection platform developed by the University at Buffalo's Multidisciplinary Digital Forensics Lab (UBMDFL). It enables users to upload images or videos and run them through multiple state-of-the-art deepfake detection algorithms simultaneously, receiving individual and aggregated authenticity scores for each piece of media. The platform aggregates a suite of AI-powered detection models trained on facial manipulation, GAN-generated imagery, and synthetic video techniques. Results are displayed with per-model confidence scores, helping users understand which aspects of the media appear manipulated and to what degree. It is designed to be accessible to non-technical users while remaining rigorous enough for researchers and forensic analysts. Developed as an academic research initiative, Deepfake-O-Meter serves journalists, researchers, educators, and digital forensics professionals who need a reliable, multi-model approach to verifying the authenticity of visual media. The tool is freely accessible via the web with no installation required.
Key Features
- Multi-Model Detection: Runs uploaded media through several independent deepfake detection algorithms at once, providing per-model scores for a comprehensive authenticity assessment.
- Image & Video Support: Accepts both image and video files, detecting facial manipulations, GAN-generated content, and synthetic media across multiple formats.
- Aggregated Confidence Scores: Presents individual model outputs alongside an aggregated score, giving users a clear signal of how likely the media has been manipulated.
- No Installation Required: Fully browser-based interface means users can analyze media without downloading software or setting up any local environment.
- Research-Grade Analysis: Built on academic research from the UB Media Forensics Lab, incorporating cutting-edge detection models used in peer-reviewed studies.
Pros
- Free & Accessible: Completely free to use with no account required, making professional-level deepfake detection accessible to anyone with a browser.
- Multiple Detection Models: Using several models simultaneously reduces blind spots that any single detector might have, improving overall reliability.
- Academic Credibility: Backed by University at Buffalo's forensics research, ensuring the underlying models are grounded in peer-reviewed science.
Cons
- Limited Throughput: As a research prototype hosted on a university server, it may have slow processing times or limited availability during high-traffic periods.
- No API or Bulk Processing: The tool is designed for individual file uploads and does not offer a programmatic API or batch processing for large-scale workflows.
- Detection Accuracy Varies: Like all deepfake detectors, performance can degrade on novel or highly advanced manipulation techniques not represented in training data.
