Google’s SynthID AI watermarking tech is being adopted by OpenAI, Nvidia, and more

谷歌的 SynthID AI 水印技术正被 OpenAI、Nvidia 等公司采用

In a few short years, we’ve gone from easily identifying AI content that featured superfluous fingers to images and videos that look shockingly realistic. How can we know what’s real in the age of AI? Google’s answer is SynthID, which it first demonstrated three years ago. The company says SynthID has since been used to label 100 billion images and videos, plus 60,000 years’ worth of audio. Those numbers are only going up now that SynthID is expanding beyond Google.

在短短几年内，我们已经从能够轻易识别出带有“多余手指”的 AI 内容，演变到面对看起来极其逼真的图像和视频。在 AI 时代，我们该如何辨别真伪？谷歌给出的答案是 SynthID，该技术于三年前首次展示。谷歌表示，自那时起，SynthID 已被用于标记 1000 亿张图像和视频，以及长达 6 万年的音频内容。随着 SynthID 走出谷歌生态，这些数字只会进一步增长。

SynthID is not Google’s only AI labeling strategy. It’s also committed to the C2PA standard, which tags content with metadata describing how it was created. Google began using C2PA more prominently with its Pixel 10 smartphones. Photos taken with the Pixel 10 include metadata describing how they were processed. If a highly zoomed image includes generative elements, it gets an AI tag, too. Google now says this same feature is coming to videos recorded on Pixel 8, 9, and 10 phones in an update in the coming weeks. It’s also adding C2PA scanning to Gemini, allowing the chatbot to explain a file’s providence based on the content labeling. This same capability will come to Chrome and Search in a few months.

SynthID 并非谷歌唯一的 AI 标记策略。谷歌还致力于 C2PA 标准，该标准通过元数据来描述内容的创建方式。谷歌从 Pixel 10 智能手机开始更显著地使用 C2PA。使用 Pixel 10 拍摄的照片包含描述其处理方式的元数据。如果一张高倍变焦的图像包含生成式元素，它也会被打上 AI 标签。谷歌现表示，在未来几周的更新中，这一功能将扩展至 Pixel 8、9 和 10 手机录制的视频。此外，谷歌正在为 Gemini 添加 C2PA 扫描功能，使聊天机器人能够根据内容标签解释文件的来源。这一功能也将在几个月内登陆 Chrome 浏览器和谷歌搜索。

That metadata is fungible, though. On the other hand, SynthID is deeply integrated with AI-generated content. The digital watermark is present in the pixels of images and videos and in the waveform of AI songs and audio overviews from products like NotebookLM. According to Google DeepMind scientist Pushmeet Kohli, the team worked hard to ensure SynthID is much harder to remove, even if you compress it, crop it, or rotate it. “A technology like this will always be attacked,” said Kohli. “There was a lot of research that we did in making SynthID robust to different kinds of transformations.”

不过，元数据是可以被篡改的。相比之下，SynthID 与 AI 生成的内容深度集成。数字水印存在于图像和视频的像素中，以及 NotebookLM 等产品生成的 AI 歌曲和音频概览的波形中。据 Google DeepMind 科学家 Pushmeet Kohli 介绍，团队付出了巨大努力，确保即使在压缩、裁剪或旋转的情况下，SynthID 也极难被移除。“像这样的技术总会受到攻击，”Kohli 说道，“我们进行了大量研究，以确保 SynthID 对各种变换具有鲁棒性。”

Last year, Google added support for SynthID detection in the Gemini app. You can upload the suspect content and ask the chatbot if it’s AI-generated. This should work reliably with all those billions of Google AI images and audio clips from the past three years. A few ambitious tinkerers have claimed to find methods for removing the hidden SynthID patterns. Google contends that none of these bypasses actually work.

去年，谷歌在 Gemini 应用中增加了对 SynthID 的检测支持。你可以上传可疑内容，并询问聊天机器人它是否由 AI 生成。对于过去三年中谷歌生成的数十亿张 AI 图像和音频片段，该功能应该能可靠地发挥作用。一些雄心勃勃的“修补者”声称找到了移除隐藏 SynthID 图案的方法，但谷歌坚称这些绕过手段均无效。

More SynthID in more places. Even if no one has been able to crack SynthID, that doesn’t matter for the vast majority of AI images on the Internet—only Google’s AI models apply SynthID. That’s going to change soon, though. Google has announced that it has partnered with several companies to add SynthID to their systems. Nvidia will implement SynthID in its Cosmos world foundation models, and OpenAI will use SynthID in its GPT 2 images. Kakao and ElevenLabs will also begin adding SynthID to their AI content.

在更多地方使用 SynthID。即使目前没有人能破解 SynthID，这对互联网上绝大多数 AI 图像来说也无关紧要——因为只有谷歌的 AI 模型应用了 SynthID。不过，这种情况很快就会改变。谷歌宣布已与多家公司合作，将 SynthID 添加到它们的系统中。Nvidia 将在其 Cosmos 世界基础模型中实现 SynthID，OpenAI 将在其 GPT-2 图像中使用 SynthID。Kakao 和 ElevenLabs 也将开始在其 AI 内容中添加 SynthID。

This doesn’t mean you’ll always be able to tell if something is AI by looking for SynthID. Plenty of publicly available models continue to produce content with no AI watermarking, and there are open models that can be trained by anyone looking to create AI images and videos on their own terms. Still, this is a step in the right direction. There will also be new paths to checking SynthID status, so you won’t even have to open Gemini just to check for the watermark. SynthID will be integrated with Circle to Search, Lens, and AI Mode. You’ll also be able to use Gemini in Chrome by sharing a tab with the content in question. You can ask any variation of “Is this AI” to get a SynthID scan with these tools.

这并不意味着你总能通过寻找 SynthID 来判断某物是否为 AI 生成。许多公开可用的模型仍在生产没有 AI 水印的内容，而且还有一些开源模型，任何人都可以训练它们来按自己的意愿创建 AI 图像和视频。尽管如此，这仍是朝着正确方向迈出的一步。未来还将有新的途径来检查 SynthID 状态，因此你甚至不必专门打开 Gemini 来检查水印。SynthID 将与“圈选搜索”（Circle to Search）、Google Lens 和 AI 模式集成。你还可以通过在 Chrome 中共享包含相关内容的标签页来使用 Gemini。你可以询问任何类似“这是 AI 生成的吗？”的问题，通过这些工具获得 SynthID 扫描结果。

There is currently not a public API for SynthID—making these scans too readily available could serve as an attack vector for those seeking to circumvent SynthID. However, Google is preparing to launch an AI content detection API as part of the company’s Gemini Enterprise Agent Platform. This will allow trusted business partners to more easily flag AI content, allowing Google to refine the API over the coming months.

目前 SynthID 还没有公开的 API——如果让这些扫描功能过于容易获取，可能会成为那些试图绕过 SynthID 的人的攻击途径。不过，谷歌正准备推出一个 AI 内容检测 API，作为其 Gemini 企业代理平台（Gemini Enterprise Agent Platform）的一部分。这将使受信任的商业合作伙伴能够更轻松地标记 AI 内容，并允许谷歌在未来几个月内不断完善该 API。