Disclaimer
OriginAI gives you a model's assessment of a video. It can be wrong in both directions, and it should never be the only thing you rely on.
Last updated October 2026
A screening tool
OriginAI uses trained models to flag videos that may have been manipulated. It isn't forensic certification, identity verification, or legal advice, and it comes with no guarantee of accuracy.
It can be wrong
A "likely real" result doesn't prove a video is genuine, and a "likely deepfake" result doesn't prove it was faked. Compression, blur, low light, a partly hidden face, editing, unusual movement, and manipulation methods the model hasn't seen can all affect the score.
The model matters
Veni HQ, Veni LQ, Vidi and Vici learned from different datasets and suit different kinds of footage. Always read a result together with the detector name, what it learned from and its threshold.
Evidence maps
An evidence map shows the parts of the face that mattered most to the score. It doesn't identify a specific edit, the tool used to make it, or who made it.
Important decisions
Don't rely on OriginAI alone for legal, law-enforcement, employment, academic, financial, medical, disciplinary or safety decisions. Check the original source, the file's history and other evidence, and ask a qualified digital-forensics professional when a lot is at stake.