Currently since few days topic of discussion has been Deepfakes. In that context, find below a compilation of definition, tips, articles and companies working on detecting Deepfakes.
As per Wikipedia, Deepfakes are – “Deepfakes (portmanteau of “deep learning” and “fake”[1]) are synthetic media[2] that have been digitally manipulated to replace one person’s likeness convincingly with that of another. Deepfakes are the manipulation of facial appearance through deep generative methods.[3] While the act of creating fake content is not new, deepfakes leverage powerful techniques from machine learning and artificial intelligence to manipulate or generate visual and audio content that can more easily deceive.[4][5]”
To quote this article: Deepfakes, explained | MIT Sloan
Here are some common tips to detect a Deepfake.
- Face — Is someone blinking too much or too little? Do their eyebrows fit their face? Is someone’s hair in the wrong spot? Does their skin look airbrushed or, conversely, are there too many wrinkles?
- Audio — Does someone’s voice not match their appearance (ex. a heavyset man with a higher-pitched feminine voice).
- Lighting — What sort of reflection, if any, are a person’s glasses giving under a light? (Deepfakes often fail to fully represent the natural physics of lighting.)
You could also check frame and pixel distortions in slow motion or via zooming in. This depends on the quality of deepfake creation software – Deepfakes can be created for both audio and video part.
Deepswap – Best Online Faceswap Tool for Face Swap Videos – There are multiple online generators as well for Deepfakes.
Long term solution as per me:
- Integrate: Neural networks-based learning for detection & Zero trust frameworks on most large tech systems like Email, Social Media, ERPs, Portals, Internet and more. This approach of zero trust is used in cyber security as well. It pushes the video/audio processor/owner/creator/manager to maintain good metadata, sources, links, published records, identity, tagging and more with hashing to confirm the authenticity and manipulation. Wikipedia and many Linux based downloads are indirectly managed this way.
Companies working on Deepfakes detection:
- https://sensity.ai/deepfake-detection/
- https://blogs.microsoft.com/on-the-issues/2020/09/01/disinformation-deepfakes-newsguard-video-authenticator/
- https://realitydefender.com/
- https://www.startus-insights.com/innovators-guide/5-top-startups-tackling-deepfakes/
- https://www.businesstoday.in/technology/news/story/startups-and-companies-that-are-taking-on-the-problem-of-deepfakes-405607-2023-11-15
- https://gizmodo.com/chatgpt-ai-12-companies-deepfake-video-image-detectors-1850480813
- https://www.intel.com/content/www/us/en/company-overview/wonderful/detect-deepfakes.html
These are not 100% perfect at detection and it’s getting harder with each passing day to manage the detection of Deepfakes as neural networks / LLMs are evolving rapidly and improving daily.
Word of caution: Before you use any online or offline Deepfake site or tool, be cognizant of the legal implications.
Email me: Neil@HarwaniSystems.in
Website: Harwani Systems (OPC) Private Limited
Blog: Innovation Ideas blog | Ideas on Innovation around Software. We Thrive On Ideas. We are Learner Centered, Open Source & Digital Focused. (techandtrain.com)
Linkedin: Neil Harwani | LinkedIn