Google Opens SynthID Detector To The Public
A watermark checker is only as useful as the number of companies that embed the mark, and Google is now routing verification for ChatGPT, NVIDIA, and Kakao content through one page while text detection stays off the table.
Reporting from 1 source: GameBusiness.jp.
Google has opened SynthID Detector, an online tool that checks images, video, and audio for SynthID, the digital watermark from Google DeepMind. It reads watermarks from Google's own AI tools and from OpenAI, NVIDIA, and Kakao, and Google says support for Apple is coming soon. It only detects the presence of a watermark and cannot judge whether a file was made by a human or an AI.
The tool reads SynthID marks that survive filters, color changes, and lossy compression, and for video and audio it shows the segments where a watermark turned up. What it will not do is broader than what it will. A file without a watermark tells you nothing, since local models and services outside SynthID leave no mark, and heavy processing with another company's AI can weaken one past detection. A detected watermark also says nothing about truth or authorship. Text is not supported, because SynthID Text works by shifting word-choice probabilities, and the mark does not survive extensive rewriting or translation.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.