We’re Making It Easier To Identify AI-generated Content Globally.

TL;DR

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Google has opened its SynthID Detector to the public worldwide in English, expanding a tool previously introduced in an early version for media professionals. The company says it can check images, video and audio for SynthID watermarks associated with Google and participating partners; it is not described as a universal detector for all AI-generated media.

Google has opened SynthID Detector to everyone worldwide in English, expanding a tool first offered in an early version to media professionals. The detector lets users check images, video and audio for SynthID watermarks associated with AI content made by Google and named partners, giving people another way to assess the origin of media they encounter online.

Google says the detector can identify content made with AI from Google or its partners, which the company names as OpenAI, NVIDIA and Kakao. Apple is expected to be added soon, according to Google; the company has not supplied a date for that addition. The announcement does not say the tool can identify AI content that lacks a compatible SynthID watermark.

The detector builds on SynthID, launched in 2023. Google says the system embeds imperceptible watermarks in images, video and audio. The company reports that it has watermarked more than 180 billion images and videos, as well as 240,000 years of audio content. Those figures are Google’s reported totals, not an independent assessment of how much AI media circulates online.

Google also points to existing verification features in Search, the Gemini app and Chrome. It says those features now handle more than one million requests daily. The announcement does not break down those requests by product or explain how they compare with use of the newly opened detector.

More People Can Check Watermarks

Opening the detector beyond media professionals could make watermark checks more accessible to people assessing images, clips or recordings shared online. That may be useful when a piece of media’s origin is uncertain, particularly as generative AI tools make it easier to produce and distribute synthetic content.

The practical reach of the change depends on which tools and formats the detector supports. Google describes checks for SynthID-marked content from itself and participating companies, not a method that can identify every AI-generated item. A result that does not find a watermark, on its own, cannot establish that media is human-made. The detector is one source of information about provenance, rather than a complete test of authenticity.

From Media Pilot to Public Access

Google introduced SynthID in 2023 as a way to add imperceptible markers to AI-generated media. The company later introduced an early version of the detector for media professionals, according to its announcement. The current release changes who can access the detector: Google says it is now available to everyone globally, with the interface in English.

The move is part of a broader set of verification features Google says it has built into Search, Gemini and Chrome. The company presents those tools and SynthID Detector as ways to provide more context about online media. The announcement does not describe a shared industry standard or say that all named partners apply the same watermark to every item generated with their products.

““Now, we’re expanding access to everyone – with the tool available globally in English starting today.””

— Google

Limits of Watermark Coverage

Google’s announcement does not specify how the detector handles media without a SynthID watermark, or whether it can assess content created using tools outside its stated partner group. It also does not provide accuracy rates, error rates, or details about how users should interpret an inconclusive result. Those limits matter because the announcement describes a watermark check, not a general-purpose way to establish whether any media was generated by AI.

The company says Apple will be added soon but gives no timing or technical details for that planned support. It is also unclear from the announcement whether the detector will expand beyond English, or whether access will change by region, file type or usage limits.

Partner Support and Language Expansion

Google says the detector is available globally in English now and that Apple support is coming soon. The company has not announced a specific date for Apple’s inclusion or shared a timetable for other languages. Further details about supported formats, coverage and result interpretation would help users understand what a detector check can and cannot establish.

For now, users can treat SynthID Detector as a way to check for compatible watermarks in supported media. Google’s announcement does not set out a broader verification standard or claim that a negative result proves a file is authentic or was made without AI.

Key Questions

What has Google announced?

Google says SynthID Detector is now available to everyone globally in English. An early version had been introduced for media professionals.

What kinds of media can it check?

Google says the tool checks images, video and audio for SynthID watermarks associated with AI content from Google and participating partners.

Does it detect every AI-generated image or recording?

The announcement does not claim universal coverage. It describes checking for SynthID watermarks, so it does not establish that content without a detected watermark was made by a person.

Which companies are included?

Google names OpenAI, NVIDIA and Kakao as partners whose AI-generated media can be checked. It says Apple support is planned soon, without giving a date.

How many items has Google watermarked?

Google reports more than 180 billion images and videos and 240,000 years of audio content watermarked to date. These are company-reported totals.

Source: rss