At Google I/O 2025, Google introduced SynthID Detector – a tool designed to identify AI-generated content with speed and precision. Developed by Google DeepMind, this tool is a major step toward ensuring transparency and trust in digital media created using Google’s AI models.
With the rise of generative AI, content authenticity is becoming harder to verify. Google’s Pushmeet Kohli, VP of Science & Strategic Initiatives at DeepMind, emphasized that SynthID Detector is built to address this challenge head-on, offering users a clear way to identify content created by AI systems like Gemini, Imagen, Veo, and Lyria.
SynthID Detector uses invisible watermarking technology, which embeds identifiers directly into AI-generated content without altering quality. These watermarks remain detectable even after the content is edited, compressed, or shared. Supported media types include Images, Audio, Video, and Text.
Detection Process:
- Upload content (image, text, audio, or video)
- The portal scans for embedded SynthID watermarks
- Results display if watermarks are found, and pinpoint exact segments or areas where they exist
For example:
- In images/videos, the tool highlights regions likely to contain AI watermarks.
- In audio, it detects specific time segments with AI signatures.
- For text, SynthID flags sections with generated content traces.
Google isn’t keeping SynthID exclusive. It has open-sourced text watermarking tools for developers to use with their own models. It has partnered with NVIDIA to integrate SynthID into Cosmos, NVIDIA’s AI video generation tool. It is teamed up with GetReal Security, a content verification platform, to detect SynthID watermarks across the web. These moves extend SynthID’s utility beyond Google’s models, supporting a broader push against AI misuse and misinformation.
SynthID is launched for early testers on 20th May 2025. A wider rollout is expected in the coming weeks. Also, video and text detection support is coming soon. Journalists, researchers, and media professionals can sign up for early access via a waitlist.