GetReal Labs

Trick or Truth week 2 results are here! And the answer is... synthetic! Our forensic models, trained to distinguish natural from AI-generated images, detect consistent signs of AI-generation. In addition, a physical analysis of the vanishing-point geometry indicates the image is inauthentic. The red lines in the annotated image correspond to parallel lines in the scene (the shelves in the background). In a natural image, these lines should converge to a single vanishing point. And lastly, although generative AI has gotten better at rendering hands, you can see a malformed finger, a tell-tale sign of generative AI. 

🎃 Trick or Truth Tuesdays returns with another spine-chilling challenge! 🎃 Is this image the real deal or a synthetic scare? Drop your guess "Real" or "Synthetic" in the comments below and see if you can spot the truth. Don’t get ghosted by a deepfake! 

🎭 Trick or Truth Results Are In! 🕵️♂️ No tricks here – it was real! Our forensic models, trained to distinguish natural from AI-generated images, detected no signs of AI-generation. In addition, a physical analysis of the vanishing-point and vanishing-line geometry point to the image being natural. The green lines in this annotated image correspond to parallel lines on a flat surface in the scene. As expected in a natural image, each set of parallel lines converge to a vanishing point. The yellow line — the vanishing line — that connects these two vanishing points corresponds to the orientation of the camera which is nearly perfectly horizontal consistent with what we see in the scene (the camera doesn’t appear to be rotated). This type of natural geometry is rarely satisfied in AI-generated images and so this adds to the evidence the image is real.

For the past 25 years, my students and I have been developing techniques to detect all forms of digital manipulation. For almost as long, I've been asked the same question "will we get to a point where you won't be able to tell if something is real or fake?" My usual response is "I don't know, but it hasn't happened yet." When Emmanuelle Saliba showed me the latest AI-powered, on-camera photo editing capabilities of the new Pixel 9, however, I started doubting my stock answer. Below are two examples where after taking a photo with the Pixel 9 I specified a region in the image and typed "add a monster" and "add a car damaged by branches." The results are pretty impressive. What makes these types of images so difficult to detect is that most of the content is natural and so models trained to distinguish natural from AI-generated images will struggle. It took us a few weeks, but we eventually noticed an interesting artifact that we can exploit to detect and localize parts of the image that were digitally modified/added. Shown to the right of each image is the result of this analysis where the blue-tinted region is identified as AI-edited. For now, I'm not going to publicize how we are doing this (no students were involved in the making of this analysis tool). Maybe this will give me a little more time to continue to say "I don't know, but it hasn't happened yet." GetReal Labs

I'm receiving a lot of questions about a video that has gone viral on X purportedly by a former student of Tim Walz. The person in the video claims to have been abused by Walz in the 90's. Based on some good investigations by several reporters, it appears that the name of the person in the video is real and corresponds to someone who went to school where Walz taught, but this person has little resemblance to the person they purport to be. Beyond the biometric inconsistencies, I am not convinced that this video is AI-generated or AI-manipulated. I analyzed three aspects of the video: (1) based on models we have trained to distinguish real from AI-generated voices, the audio shows only minor evidence of being AI-generated but not overwhelmingly so. (2) based on models we have trained to detect real from AI-generated/manipulated faces, the face shows no evidence of being AI-generated. (3) based on a model we have built that measures the synchronization between the mouth movement and the spoken word, I find no evidence of a lip-sync deepfake. (4) And finally, there are none of the usual visual signs of this being a deepfake video in terms of things like inconsistencies between the head movements and spoken word. Some of the artifacts in the mouth and teeth can, I believe, be explained by video compression artifacts. In summary, I think it is more likely that this is a cheapfake in which someone is simply impersonating another person. GetReal Labs UPDATE: Fact check from AFP https://lnkd.in/g5HRCwby UPDATE: Russian Propaganda Unit Appears to Be Behind Spread of False Tim Walz Sexual Abuse Claims https://lnkd.in/gYW75W7r UPDATE: U.S. intel officials say Russia is behind attempts to smear Tim Walz https://lnkd.in/gFc_eRVf

🚨 In case you missed it! 🚨 A quick reminder that GetReal Labs' CEO, Matthew Moynahan, will be speaking at the 6th edition of the Transatlantic Leaders Forum (TLF) on the 18th, hosted by FrenchFounders and co-organized with Tikehau Capital & Goldman Sachs. Mark your calendars for insights on deepfakes and digital integrity!

🎃 Introducing Trick or Truth Tuesdays! 🎃 As we move closer to Halloween and get into the fall spirit, we’re kicking off a spooky new series! Each week, you get to decide: Are these images real or synthetic? Test your skills and comment below: Real or Synthetic?

For anyone in New York, you don't want to miss our fearless leader speaking about his insights in the cybersecurity realm. Get your ticket here below!

Can your team differentiate between a genuine conversation and a deepfake-driven deception? A U.S. Senator was recently deceived by a deepfake impersonating a Ukrainian official, underscoring the growing threat of AI-powered impersonations. As these technologies evolve, so must our vigilance and security measures. Don’t wait until it’s too late.

We are thrilled to announce that GetReal Labs’ CEO, Matthew Moynahan will be joining the 6th edition of the Transatlantic Leaders Forum (TLF), hosted by Frenchfounders and co-organized with Tikehau Capital & Goldman Sachs. Matt will be discussing the growing challenges of malicious manipulated content and the evolving role of cybersecurity in today’s digital landscape. As CEO of GetReal Labs, the leading authority on deepfakes, Matt has focused on anticipating industry shifts and positioning companies for success. The forum’s theme DARE, CARE, SHARE, aligns perfectly with our mission to protect digital integrity. We look forward to Matt sharing insights and connecting with global leaders. 👉 Join the discussion on October 18th in New York, get your ticket here: https://lnkd.in/dxY_ygXz #TLF24 #Cybersecurity #Deepfakes #Innovation #Leadership