水印与内容凭证Watermarks and Content Credentials
Watermarks and Content Credentials
识别 AI 生成内容的两条路：嵌进内容的水印，和记录出处的元数据。Two routes to identify AI-generated content: watermarks embedded in the content, provenance metadata beside it.
DeepMind 的 SynthID 把不可见数字水印直接嵌入生成内容本身。图像、音频、文本、视频都覆盖。
水印人眼不可察觉，但可被检测技术识别。
C2PA「内容凭证」走另一条路：用元数据记录内容的出处与编辑历史。
两者互补：一个改内容本身，一个改内容的说明书。
DeepMind's SynthID embeds invisible digital watermarks directly into generated images, audio, text, or video.
The watermarks are imperceptible to humans but detectable by matching technology.
C2PA 'Content Credentials' takes the other route: metadata recording the content's origin and edit history.
They complement each other: one changes the content itself, one changes its paperwork.
「这段内容是不是 AI 做的」目前没有万能答案。知道两条技术路线各自的强弱，才不会被「AI 检测器」的绝对承诺骗到。'Is this AI-made?' has no universal answer yet. Knowing each route's strengths keeps you from being sold absolute promises by 'AI detectors'.