Google于近日推出SynthID网站工具，允许用户在线验证图像、视频或音频是否由AI生成13。这一工具此前仅限记者和研究人员通过Google I/O获得测试机会，现已向全球用户开放1。用户可通过SynthID.com网站检测内容，无需借助Gemini模型查询3。
该工具支持多种常见媒体格式，包括JPG、JPEG、PNG、BMP、WEBP、AVIF、HEIC、HEIF、TIFF、GIF等图像格式，MP4、MOV、WEBM等视频格式，以及WAV、MP3、OGG、FLAC、AAC、M4A等音频格式1。SynthID的检测原理是识别编码在像素和音频波形中的隐形信号3。Google在2023年引入了这一技术，并在旗下多个AI模型和生成工具中应用水印，包括Gemini、Veo、Lyria等1。该工具已在Gemini应用和Google Chrome浏览器中集成验证功能1。
SynthID已成为行业合作标准13。OpenAI、Nvidia、Kakao等多家公司支持该技术，Apple也计划在近期加入1。改进后的检测器现已支持来自ChatGPT等合作伙伴的水印，不仅限于Google版本3。该平台目前每天收到100万次内容验证请求1。不过，业界也存在其他水印标准——Microsoft和Meta各自拥有独立的水印和验证方案1。
Google has rolled out a publicly accessible SynthID website that enables users to detect whether images, videos, or audio files were created by artificial intelligence 13. The tool, which was previously restricted to journalists and researchers during testing at Google I/O, now operates as a standalone detector at SynthID.com and supports content from multiple AI companies beyond Google itself 13.
The detector accepts various media formats, including JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, and GIF images; MP4 and WEBM videos; and WAV, MP3, OGG, FLAC, AAC, and M4A audio files 1. SynthID embeds invisible signals directly into pixels and audio waveforms to mark AI-generated content 3. Google's AI models—including Nano Banana, Veo, Lyria, and generative tools like Gemini, Flow, ProducerAI, and Vids—all utilize SynthID watermarking 1. The technology has also been adopted by partners including OpenAI, Nvidia, and Kakao, with Apple planning to implement it soon 1. Google has additionally integrated SynthID verification into its Gemini application and Google Chrome browser 1.
The detector has been integrated with over 180 billion images and videos plus 240,000 years of audio content 3, and currently receives approximately one million content verification requests daily 1. While SynthID represents a significant step in identifying synthetic media, other technology companies maintain alternative approaches: Microsoft and Meta have developed their own watermarking and verification standards, though these tools are not entirely reliable 1.
评论
还没有评论，欢迎留下第一条。