谷歌已向所有人开放 SynthID 检测器。这个免费网站用于检查图像、视频或音频文件是否带有谷歌的 SynthID 水印——这是其 AI 工具为其创建的内容添加的一种隐藏信号。谷歌 DeepMind 的普什米特·科利（Pushmeet Kohli）在公司博客文章中写道，该工具已于周三在全球范围推出英文版。
Google has opened its SynthID Detector to everyone. The free website checks whether an image, video or audio file carries Google’s SynthID watermark, a hidden signal its AI tools add to what they create. The tool is available globally in English from Wednesday, Google DeepMind’s Pushmeet Kohli wrote in a company blog post.
它可以标记由谷歌及其合作伙伴 OpenAI、英伟达和 Kakao 的 AI 制作的媒体。谷歌表示，对苹果的支持将“很快”推出。
It flags media made with AI from Google and its partners OpenAI, Nvidia and Kakao. Support for Apple is coming “soon”, Google said.
科利写道：“随着 AI 工具变得更好、更容易使用，拥有关于你在网上遇到的内容的清晰背景，有助于你对信任什么做出明智的决定。”
“As AI tools become better and easier to use, having clear context about the content you encounter online helps you make informed decisions about what to trust,” Kohli wrote.
从记者到所有人：谷歌于 2025 年 5 月推出了检测器的早期版本，供记者、媒体专业人士和研究人员使用。该公司表示，自 2023 年以来，已为超过 1800 亿张图片和视频以及相当于 24 万年长度的音频打上了 SynthID 水印。目前其在搜索、Gemini 应用和 Chrome 中的检查请求每天已处理超过一百万次。
From journalists to everyone Google introduced an early version of the detector in May 2025, for journalists, media professionals and researchers. It has watermarked more than 180 billion images and videos with SynthID since 2023, along with 240,000 years of audio, the company said. Its checks in Search, the Gemini app and Chrome now handle more than one million requests a day.
据 Engadget 报道，使用该网站需要登录谷歌、OpenAI 或苹果账户。
Using the site requiresg in with a Google, OpenAI or Apple account, Engadget reported.
该公司在当时发布时表示，SynthID 会标记谷歌 Gemini、Imagen、Lyria 和 Veo 模型生成的内容。OpenAI 于 5 月采用了 SynthID。它还运营自己的验证页面，用于检查图像和音频中的 SynthID 和 C2PA 信号。
SynthID marks content from Google’s Gemini, Imagen, Lyria and Veo models, the company said at that launch. OpenAI adopted SynthID in May. It also runs its own verification page, which checks images and audio for SynthID and C2PA signals.
它无法告知你的信息：该检测器仅查找 SynthID 水印。其他公司使用不同的标准，因此它无法捕获所有 AI 生成的内容。Engadget 报道称，在某些情况下，水印也可能会被剥离。
What it cannot tell you The detector only finds the SynthID watermark. Other companies use different standards, so it cannot catch all AI-generated content. Watermarks can also be stripped in some cases, Engadget reported.
检测结果也无法说明文件是完全由 AI 制作的还是仅经过 AI 编辑。谷歌的示例结果显示，该媒体“由谷歌 AI 制作或编辑”。它还警告称，自那以后该文件可能已被修改。
A result also does not say whether a file was made entirely with AI or only edited with it. Google’s sample result says the media “was made or edited with Google AI”. It also warns that the file may have changed since.
Meta 运营着自己的检查器，用于检测其 Muse 图像模型的图像。今年 7 月，路透社发现 Meta 的工具未能识别出其自身被裁剪的造假图像。本周，OpenAI 还开始在欧盟为 ChatGPT 文本添加水印。
Meta runs its own checker for images from its Muse Image models. In July, Reuters found that Meta’s tool missed its own cropped fakes. This week, OpenAI also began watermarking ChatGPT text in the EU.