- Home
- AI Deepfake Detection Tools
- DeepFake-O-Meter
DeepFake-O-Meter
Buffalo, New York, United States
What it is
DeepFake-O-Meter is a free web platform from the University at Buffalo’s UB Media Forensics Lab for assessing whether images, videos, and audio may be AI-generated. It combines results from 37 integrated forensic research models. Users can choose detection algorithms using listed measures such as accuracy, running time, and year developed. Each selected algorithm returns a percentage indicating the likelihood that submitted content was AI-generated. The catalog includes image, video, and audio methods, and the project is described as open source with algorithm source code publicly accessible. Before uploading, users are asked whether they want to share media with researchers; the team says shared submissions can support training detection algorithms. Analysis comes from multiple methods rather than a definitive authenticity decision. The web service offers account registration with an email, username, and password.
For and against
For
- Free web access
- Analyzes images, video, and audio
- 37 forensic research models
- Algorithm-selection metrics listed
- Open-source algorithm code
Against
- Results are not definitive authenticity judgments
- WAV2LIP-STA may suffer under heavy compression
- Media sharing with researchers is requested before upload
The verdict
DeepFake-O-Meter brings many research detection methods into one browser-based service. Its percentage outputs and interpretation caveat make it an analysis aid rather than a final authenticity ruling.
Who it suits
It suits social media users, journalists, and law enforcement seeking multiple analyses of potentially synthetic media. Users who need to make their own authenticity judgment may appreciate its non-definitive interpretation.
On the record
- Account access
- The site provides account sign-up and login pages, and its login page offers a guest option to return to the home page.zinc.cse.buffalo.edu · read 7 Oct 2026
- Current use
- The home page reports that Deepfake-o-Meter was used on the AFP fact check website to detect a deepfake video.zinc.cse.buffalo.edu · read 7 Oct 2026
- Detection models
- The models page lists 37 integrated forensic research models.zinc.cse.buffalo.edu · read 7 Oct 2026
- Intended users
- The University at Buffalo describes the platform as serving users including social media users, journalists, and law enforcement.buffalo.edu · read 7 Oct 2026
- Interpretation caveat
- The platform provides analysis from multiple methods and does not make strong claims about whether uploaded content is authentic.buffalo.edu · read 7 Oct 2026
- Maker
- The site attributes DeepFake-O-Meter to the University at Buffalo and the UB Media Forensics Lab.zinc.cse.buffalo.edu · read 7 Oct 2026
- Media types
- The platform analyzes images, videos, and audio.buffalo.edu · read 7 Oct 2026
- Model categories
- The model catalog includes methods labeled for image, video, and audio detection.zinc.cse.buffalo.edu · read 7 Oct 2026
- Model limitations
- The model catalog notes that WAV2LIP-STA performance may suffer under heavy compression.zinc.cse.buffalo.edu · read 7 Oct 2026
- Model selection
- Users can select detection algorithms using listed metrics that include accuracy, running time, and year developed.buffalo.edu · read 7 Oct 2026
- Multiple analyses
- The University at Buffalo says users can run multiple detection algorithms, each returning a percentage likelihood that content was AI generated.buffalo.edu · read 7 Oct 2026
- Open source
- The platform is described as open source, with algorithm source code publicly accessible.buffalo.edu · read 7 Oct 2026
- Project team
- The site identifies the UB Media Forensics Lab as the developer and lists Siwei Lyu as project director.zinc.cse.buffalo.edu · read 7 Oct 2026
- Purpose
- DeepFake-o-Meter aggregates results from research AI media detection models to support media authenticity assessment.zinc.cse.buffalo.edu · read 7 Oct 2026
- Research models
- The detection models page lists 37 integrated forensic research models.zinc.cse.buffalo.edu · read 7 Oct 2026
- Result interpretation
- The University at Buffalo says the platform provides analysis from a broad range of methods and leaves users to decide whether content is real.buffalo.edu · read 7 Oct 2026
- Results
- Each selected algorithm returns a percentage representing the likelihood that content was AI-generated.buffalo.edu · read 7 Oct 2026
- Submission sharing
- Before upload, users are asked whether they want to share the media with researchers; the team says shared submissions can help train detection algorithms.buffalo.edu · read 7 Oct 2026
