Deepfake Alert: How to detect AI-generated images and videos using Gemini AI app

For content produced by non-Google AI sources, Gemini will be able to provide general observations about common AI attributes such as unusual lighting, repetitive patterns, and errors in small details. With such results, users can make an educated guess on whether the image or a video is a deepfake or not.

Earlier this month, we saw cyber criminals reportedly using a deepfake video of Sudha Murthy to lure potential victims to a fake investment scam. Before it could harm people, the fake video racket got busted by news websites in the nick of time.

However, for threat actors, one failure in their mission doesn't deter them, and they don't stop working. They use generative Artificial Intelligence (gen AI)-powered tools to create more fake images and videos of other high-profile people and come up with new nefarious ways to hoodwink naive online users.

To curb such cyber crimes, search engine Google has come up with tools on the Gemini AI app to validate whether an image or video is synthetic or not.

Google's SynthID tool

Here's how to detect AI generated images and videos using Google Gemini AI app:

Step1: Open the Gemini app or visit gemini.google.com on a computer.

[Note: User must have a Gmail account to use Gemini app]

Step 2: Upload the content by tapping the "Add files" icon.

For images, a single file can be uploaded (under 100MB).

For videos, the file must be under 100MB and no longer than 90 seconds.

Step 3: Once the file is uploaded, ask a verification question in the chat box, such as:

"Was this image/video created or edited by Google AI?" "Is this AI-generated?"

Step 3: Google Gemini will scan for the SynthID watermark in the visual and, for videos, audio tracks.

For the uninitiated, SynthID is developed by Google's DeepMind division. As noted earlier, it works on gen AI language models for Gemini AI. It comes with several nuanced technologies to ensure the artificial images are aesthetically good and not marred with noticeable marks that compromise visual beauty.

The SynthID tool will ensure that even if the AI-generated image is edited with more layers of filters using other applications, will still not get erased or hide the digital watermark. The SynthID is deeply ingrained with the metadata of the photo and it cannot be erased.

However, for content from non-Google AI sources, Gemini will only be able to provide general observations about common AI attributes such as unusual lighting, repetitive patterns, and errors in small details. With such results, users can make an educated guess on whether the image or a video is deepfake or not.

Get the latest news on new launches, gadget reviews, apps, cybersecurity, and more on personal technology only on DH Tech