Gemini App Can Now Identify AI Deepfake Videos: Here's How To Use

Hyderabad: Google DeepMind, the tech giant’s AI research wing, has integrated the SynthID Detection Tool in the Gemini app, allowing users to detect AI deepfake videos generated using Google's video generation tools like Veo and Flow. Earlier, this tool existed on a dedicated portal, allowing users to check for an invisible watermark known as SynthID in text, audio, video, and images, generated by Google's AI models. Now, the tool is part of the Gemini app and accessible to everyone.

With this, Google aims to provide users with greater transparency by distinguishing between human-made and AI-generated content, reducing confusion and curbing misinformation.

New Content Transparency Tools

To determine whether content is human-made or AI-generated, users in the Gemini app can simply upload a video and ask if the image or video was generated using Google AI. Gemini will scan for the invisible SynthID watermark across both audio and visual tracks. Sythn ID helps Gemini to identify whether the content is AI-generated.

Gemini then provides context for the verified content and specifies all AI elements generated using Google AI.

The new content transparency tools will display a message such as “SynthID detected within the audio between 10-20 secs.” if any SynthID-tagged video or image is found, or “No SynthID detected in the visuals,” if none is present.

Users can upload files up to 100MB in size and 90 seconds in length.

Earlier, Google had integrated the AI image verification tool into the Gemini app. With the addition of AI video verification, Gemini can now scan and verify both images and videos + audio in uploaded content. The verification can be performed in all languages and countries where the Gemini app is supported.

SynthID

Google DeepMind introduced SynthID on August 29, 2023, as an invisible watermarking tool for identifying AI-generated content, including audio, images, text, and videos. SynthID is embedded across all products created using Google’s generative AI consumer products. DeepMind claims that SynthID does not change the image or video quality. It is designed to withstand modifications like cropping, adding filters, changing frame rates, or lossy compression.

Earlier this year, DeepMind rolled out SynthID Detector, which enables users to recognise AI content using Google tools by detecting the SynthID watermark. This tool helps users identify images made by Imagen, videos generated by Veo, audio produced by Lyria, and text generated by Gemini.

However, according to a study conducted by researchers at the University of Waterloo’s Cybersecurity and Privacy Institute, the watermark may not be foolproof. They developed a tool named UnMarker, which could destroy digital watermarks like Google’s SynthID and Meta’s Stable Signature without having knowledge of the watermarking mechanism to remove these watermarks.