As AI-generated images and videos rapidly spread across social media and messaging platforms, it is becoming increasingly difficult to distinguish between authentic content and manipulated media. While visible watermarks provide one clue, many AI-generated visuals contain invisible identifiers that are difficult to detect. According to company information, Google is now offering users a practical way to verify such content through its Gemini AI model, which can identify hidden AI watermarks embedded in media files.
Google is making Gemini available in products like Gmail and Docs, positioning the tool as a way to help users distinguish between authentic and AI-generated content using the latest watermarking technology being adopted across the industry. The company says Gemini can analyze uploaded images and videos to detect SynthID, a digital watermark designed specifically for AI-generated media.
To check if a photo or video is generated by AI, users need to open the Gemini app on their phone and upload the media file they want to check. Google has informed that video uploads cannot exceed 100MB in size and cannot exceed 90 seconds. After uploading, users can ask Gemini to verify their content by filling in prompts such as whether the media was generated using Google AI or whether it was generated by AI using some tool.
Gemini then scans the content for the presence of a SynthID watermark, which can be embedded in both the audio and visual layers of AI-generated media. Google said Gemini is trained to analyze these markers and use AI to identify which parts of the video or image were generated and when the AI-generated elements are displayed.
According to Google, SynthID watermarks are designed to be detected only by AI systems and related tools, are invisible to the human eye, and difficult to remove without degrading the content. The company launched SynthID in 2023 and announced that since then, more than 20 billion pieces of AI-generated content have been watermarked using the technology.
Google also said it is working with third-party companies to extend the SynthID watermark to content generated by other AI tools, with the goal of improving transparency and enabling more reliable identification of AI-manipulated media across its platforms.
First publication date
