Google launches SynthID Detector to spot AI-generated content across text, audio, and video

Generative AI is getting so good that spotting synthetic audio, video, or images with the naked eye has become nearly impossible. Big tech companies have promised invisible watermarks for years, but here is the catch: if nobody outside their walls can actually read those watermarks, they are essentially useless.

Google just launched SynthID Detector, a centralized web portal designed to verify whether media was created using Google's generative AI tools.

One detector for four modalities

SynthID originally started as an imperceptible watermarking system for AI-generated images. Over time, Google expanded the technology across text, audio, and video, baking it into flagship models like Gemini, Imagen, Lyria, and Veo. According to Google, over 10 billion pieces of content have already been watermarked.

The new SynthID Detector brings all those formats under one roof. Users can upload an image, audio clip, video, or block of text to scan for SynthID watermarks.

Instead of just returning a binary pass or fail, the portal highlights the exact areas where watermarks are detected. For instance, it pinpoints specific timestamps in an audio track or highlights suspicious regions inside an image.

Google is rolling out access to early testers first, with journalists, media professionals, and researchers invited to join a waitlist.

Expanding beyond Google’s walled garden

A detection system that only works inside Google's ecosystem would have a hard ceiling. To combat this, Google previously open-sourced its text watermarking technique so independent developers could adopt it.

Earlier this year, Google also partnered with NVIDIA to embed SynthID into videos generated by NVIDIA's Cosmos preview NIM microservice. Alongside the portal launch, Google announced a partnership with content verification platform GetReal Security to integrate SynthID detection into third-party workflows.

Why it matters

We think launching a public verification portal is a smart, necessary step, but it is critical to understand what this tool actually is: a watermark reader, not a universal deepfake detector.

For investigative journalists and fact-checkers, this provides a tangible, authoritative way to verify Google-originated synthetic media during high-stakes news events. Giving third-party platforms like GetReal Security detection access also prevents it from feeling entirely like a captive proprietary feature.

However, the obvious limitation is coverage. The detector cannot identify AI content generated by OpenAI, Anthropic, Midjourney, or open-source models unless they adopt the same standard. If a bad actor uses an unwatermarked open model to spread disinformation, SynthID Detector will come up blank.

It looks like the tech industry is inching closer to accountability, but until universal watermarking standards are universally enforced, verification will remain fragmented across competing walled gardens.