Google Opens SynthID Detector for AI Images, Video and Audio Across Supported Platforms

Google has launched SynthID Detector, a public website that checks images, video and audio for embedded AI watermarks. The service expands access beyond the journalists, media professionals and researchers who tested it after Google I/O in 2025, giving anyone with a supported account a way to inspect digital media. TechCrunch reported the public launch on October 7, 2026, while Google’s AI Blog and DeepMind describe SynthID as part of a broader effort to show how content was created and edited.

The detector identifies watermarks rather than judging whether a file simply looks artificial. That distinction matters because generative models can produce realistic media that defeats visual inspection, while watermark detection depends on information embedded during creation. Google’s public SynthID pages and developer documentation frame the system as a provenance tool covering multiple media types.

SynthID Detector analyzes submitted files for detectable watermarks associated with participating AI systems. The service covers still images, video and audio, according to TechCrunch, and reports whether it finds a watermark from one of its supported companies. Ars Technica said the detector can recognize watermarks from Google, OpenAI and Apple accounts, although the available excerpt does not establish that each company’s generation tools are covered equally.

The technology is designed around provenance signals rather than a universal AI classifier. Google’s developer material describes SynthID as a set of watermarking and detection tools, including methods for content generated by language models. DeepMind’s SynthID project provides the research and product foundation, while the public website turns that work into a consumer-facing check.

The website is open to the public, but users must log in before submitting content. Ars Technica reported that Google, OpenAI and Apple accounts are accepted, a choice that lowers the need to create another credential while still putting a gate in front of large-scale anonymous use. That requirement also gives Google a measure of control over demand and abuse as the service becomes widely available.

The login condition creates a practical tradeoff for journalists, researchers and ordinary users checking material quickly. A browser tool that accepts common accounts is easier to use than a specialist forensic workflow, but it remains less frictionless than a public upload endpoint. Google’s public documentation places SynthID within responsible AI tooling, while the company’s blog presents provenance checks as one part of a larger system for understanding media history.

SynthID can identify a watermark only when the relevant generation system embedded one and the signal survives later handling. That makes the detector useful for supported content, but it doesn't establish that media without a detected watermark was created by a person. Re-encoding, editing, screenshots, cropping or other transformations can also affect detection, depending on how the watermark was designed.

The distinction separates provenance from the broad promise of an AI detector. Originality’s coverage focuses on Google’s watermark becoming publicly accessible, while the technical discussion around SynthID centers on embedding signals that remain detectable through common changes. OpenAI’s provenance work, listed among the related material, reflects the wider industry push toward standardized ways to identify how media was produced.

Google’s launch puts a major platform behind a consumer-facing provenance workflow at a time when synthetic images, cloned voices and generated video are spreading across social networks, newsrooms and commercial services. The immediate value is practical: users can test suspicious files without installing specialist software, and publishers can add another check before distributing material.

The larger test is interoperability. A detector becomes more useful when platforms, model makers and editing tools preserve compatible signals across the media pipeline. Google’s AI Blog, DeepMind documentation and OpenAI’s provenance work point toward an ecosystem in which creation and editing history travels with a file, though the excerpts provide no evidence of a universal standard or complete coverage across providers.

SynthID’s public release gives users an immediate way to check supported images, videos and audio, but its result should be treated as one piece of evidence. A positive match can connect content to a watermarking system, while an inconclusive result cannot settle who made the file or whether it was edited after creation.

Google’s next challenge is coverage, resilience and trust. Wider support for media generators, clearer result explanations and less friction around access would make the service more useful to news organizations, researchers and consumers. The launch also gives other AI companies a visible benchmark: watermarking must work outside demonstrations, survive ordinary distribution and remain understandable to people who aren't forensic specialists.