Google DeepMind unveiled the SynthID Detector portal at I/O, a single‑stop tool that scans images, audio, video, and text for watermarks embedded by Google's AI models. The system flags the exact segments most likely to carry a SynthID tag, giving instant proof of origin. With over 10 billion pieces of content already watermarked, the service promises quick verification for developers and partners.

Initially launched for imagery, SynthID now marks text, audio, and video generated by models such as Gemini, Imagen, Lyria, and Veo. The watermark remains intact through compression, sharing, and other transformations, ensuring traceability. DeepMind also open‑sourced the text‑watermarking code, inviting external developers to embed the same protection in their own AI pipelines and help maintain integrity worldwide content economy today.

To broaden reach, DeepMind partnered with NVIDIA, embedding SynthID into videos from the NVIDIA Cosmos™ preview NIM microservice. Simultaneously, the company joined forces with GetReal Security, the leading content‑verification platform, to enable third‑party detection of SynthID tags. Early testers, including journalists and researchers, can now join a waitlist, while the public rollout remains pending for all users in 2024 soon.