Google introduced a new website on Tuesday designed to help general internet users verify whether an image, audio clip, or video was produced using artificial intelligence. The verification utility was initially rolled out to a limited group of journalists, researchers, and media professionals during Google I/O last year for evaluation, and the company has now made public access available to everyone across the globe.
Broad Format Compatibility for Multimedia Content
To ensure practical utility across various media files, the platform accepts an extensive assortment of file extensions. For still images, the site handles JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, TIF, and GIF formats. Users seeking to examine moving visuals can upload MP4, MOV, and WEBM video recordings. Furthermore, audio verification encompasses WAV, MP3, OGG, FLAC, AAC, and M4A sound tracks, allowing people to cross-check almost every common media format encountered online.
The Mechanics of SynthID and Industry Ecosystem
The verification mechanism relies on SynthID, an imperceptible watermarking system that Google introduced back in 2023. Generative models built by the company, including Nano Banana, Veo, and Lyria, alongside creation products like Gemini, Flow, ProducerAI, and Vids, automatically apply SynthID watermarks to their generated output. Beyond Google's own tools, companies such as OpenAI, Nvidia, and Kakao also support SynthID, with OpenAI maintaining its own verification site for media inspection, while Apple is said to be preparing support soon.
App Integration and Watermark Reliability Limits
Google has incorporated direct SynthID verification within Google Chrome and the Gemini application. According to figures shared by the company, users currently submit 1 million content verification requests every single day. Elsewhere in the tech sector, companies like Microsoft and Meta have developed proprietary frameworks for embedding digital watermarks and auditing media assets. Nevertheless, these scanning systems remain imperfect, at times failing to accurately detect generated files produced even by their creators' own underlying models.
