SynthID Detector was opened to everyone by Google on October 7, 2026. For more than a year, the verification portal had been limited to invited journalists, media professionals, and researchers. Now anyone can visit the site, upload an image, a video, or an audio clip, and check whether it carries a hidden SynthID watermark.
What the upload actually checks
SynthID is a watermarking technology Google DeepMind introduced in 2023. The signal is embedded at the moment content is generated, invisible to the eye and inaudible in audio, and it usually survives compression, cropping, and re-saving. The Detector does not guess at visual style; it reads back that pre-embedded mark and highlights which regions or segments of a file most likely contain it. Content produced by Google tools such as Gemini, Nano Banana, Veo, and Lyria is watermarked this way, and OpenAI, NVIDIA, and Kakao also support SynthID, so coverage extends beyond Google's own products.
No watermark found does not mean human-made
This is the easiest point to misunderstand. The Detector can only recognize marks from within the SynthID system. An image made by a model that never adopted SynthID, or one processed heavily enough to damage the mark, will come back clean. Microsoft and Meta run their own watermarking and provenance standards, which are not interchangeable, and evaluations have found that vendor tools sometimes miss content from their own models. The safer way to use it is as one piece of evidence rather than a final verdict: a found watermark all but confirms the source, while a negative result only means this particular trail went cold.
Why Google is widening the door
Google says about one million verification requests are already made through its products every day, and SynthID checks are also built into the Gemini app and Chrome. Opening a standalone site to the public targets the most common everyday moment: pausing before sharing, quoting, or paying for a suspicious file and checking it first. For the content industry, watermark detection moving from professional verification desks to a public entrance turns provenance into infrastructure rather than a bonus feature. Its ceiling is equally clear: detection only becomes powerful when the generation side adopts shared marking at scale.
