Google Opens SynthID Detector Globally With Partner AI Content Checks

Google opened its SynthID Detector to everyone globally in English on October 7, 2026, when an announcement on The Keyword said the portal at synthid.com now checks images, video, and audio for AI watermarks from Google and partners OpenAI, NVIDIA, and Kakao, with Apple to follow.

The post, written by Pushmeet Kohli, VP of Science and Strategic Initiatives at Google DeepMind, follows an early version of the Detector that Google introduced in 2025 for media professionals. The SynthID Detector site is explicit about its scope, stating in its FAQ: “This is not a general AI detector.” It can identify only media from companies that have adopted SynthID technology, and it says work continues on extending partnerships to cover more generated media.

Watermarking Scale and Built-In Verification

Google reported that since launching SynthID in 2023, it has watermarked more than 180 billion images and videos, along with 240,000 years of audio content. The company said the public Detector joins built-in verification features in Search, the Gemini app, and Chrome, which it said now regularly handle over 1 million requests daily.

Those built-in features were documented earlier in 2026. In a May 19, 2026 post, Google said SynthID verification for image, video, and audio in the Gemini app had been used 50 million times globally, with the capability expanding to Search that day and to Chrome over the following weeks. That post reported more than 100 billion watermarked images and videos and 60,000 years of audio.

How the Watermarks Work

Google DeepMind’s SynthID page describes watermarks embedded directly into AI-generated images, audio, text, and video, imperceptible to people while remaining detectable by SynthID’s technology. For images and video, the watermark is added at the moment of creation and is designed to remain detectable after modifications such as cropping, added filters, frame-rate changes, and lossy compression. The audio watermark, applied to output from Google’s Lyria music-generation model and NotebookLM’s podcast feature, is described as inaudible and resistant to added noise, MP3 compression, and changes in playback speed. For text from the Gemini app, SynthID adjusts the probability scores a language model assigns to each candidate word, forming a watermark that does not alter the visible output.

The Detector portal states that the watermark sits inside the content itself, in the pixels of images and video and the frequency components of audio, rather than in metadata that can be lost. It analyzes three content types: images, where it checks for a watermark’s presence, and video and audio, where it identifies watermarked segments. Supported formats include JPG, PNG, WEBP, and HEIC among other image types; MP4, MOV, and WEBM video; and WAV, MP3, OGG, FLAC, AAC, and M4A audio.

Stated Limits and Usage Terms

The Detector returns a result of watermark detected or watermark not detected. The site cautions that a negative result does not rule out AI generation, since the content may have come from a model without SynthID, or heavy modification may have degraded the signal. The FAQ also notes rare cases in which the system can falsely trigger on content that carries no watermark. It recommends uploading the highest-quality file available and gathering multiple points of data before reaching a conclusion, and it presents SynthID as a single component of a wider content-provenance strategy rather than a guarantee on its own.

The portal’s additional terms of service prohibit reverse-engineering or attacking the detection logic, automating or scraping queries, and creating multiple accounts or otherwise evading usage quotas, with Google reserving the right to suspend access. Under the site’s privacy notice, uploaded media is processed in real time and deleted immediately after results are returned, while a digital signature of the media is kept for 24 hours to enforce daily quotas. Signing in through a third-party account, such as Apple, is supported.

From Limited Test to Public Tool

Google first announced the SynthID Detector on May 20, 2025, at its I/O developer conference as a verification portal for content made with its own AI tools. That version marked the audio segments where it found a watermark and the regions of an image most likely to contain one. Access initially went to early testers, including journalists, media professionals, and researchers, through a waitlist, and Google said more than 10 billion pieces of content had been watermarked at the time.

The 2025 post also said SynthID had expanded beyond imagery to text, audio, and video across Google’s Gemini, Imagen, Lyria, and Veo models; that the company had open-sourced SynthID text watermarking; and that NVIDIA had partnered to watermark video generated by its Cosmos models. It also disclosed a partnership with content-verification firm GetReal Security.

By May 2026, Google said OpenAI, Kakao, and ElevenLabs were extending SynthID watermarking across more of their AI-generated content, and it had launched an AI Content Detection API on Google Cloud’s Gemini Enterprise Agent Platform for business users. The October 7, 2026 announcement names Apple as a forthcoming partner, saying support for detecting content made with its AI tools is coming soon.