ThreadCulture, media and the new divides

AI & Tech · Mountain View and London

Google opens SynthID Detector to the public to verify AI-generated images, audio, and video

Google made its SynthID Detector tool freely available on Wednesday, allowing users to check whether images, audio, and video contain digital watermarks embedded by AI tools from Google, OpenAI, Nvidia, and Kakao.

AI text based on reports by ABC TU DIARIO EN ESPAÑOL, Notícias ao Minuto, Deutsche Welle and other publishers. Published . How it is made

For membersThe first 30 seconds for everyone. Synthetic voice (AI).

In brief

Mountain View, London · United States, United Kingdom
How it came to this
  1. 2023Google launches SynthID watermarking for AI-generated images, video, and audio
  2. May 2025Google releases detector to journalists and announces Search and Chrome integration
  3. 7 OctGoogle opens SynthID Detector globally to the general public at synthid.com

Public launch of the detection platform

Google made its SynthID Detector freely accessible to the public on Wednesday through the synthid.com website. The web platform allows any user to upload an image, audio file, or video to determine if it carries an invisible digital watermark embedded during file generation. Previously, access to the tool had been limited to media professionals, journalists, and researchers following an initial release in May 2025. The detector is currently available globally in English and requires no subscription fees or specialized technical expertise. Users can log in to the portal using standard credentials from Google, Apple, or OpenAI accounts before submitting media for automated analysis.

Cross-industry watermarking coalition

The system identifies content generated across Google's own AI model family, including the Gemini multimodal model, the Imagen image generator, the Lyria music model, and the Veo video generator. In addition to Google's proprietary tools, the platform detects watermarks from industry partners including OpenAI, Nvidia, and South Korean tech firm Kakao. Google confirmed that Apple will also integrate into the verification network in the near future. The participating companies use SynthID to insert imperceptible data patterns directly into content without degrading the visual or audio quality for human viewers. Pushmeet Kohli of Google DeepMind outlined the purpose of the public release in a company announcement.

“As AI tools become better and easier to use, having clear context about the content you encounter online helps you make informed decisions about what to trust.”

Pushmeet Kohli, vice President of Research at Google DeepMind

Implementation scale and verification traffic

Google first introduced the SynthID watermarking technology in 2023. Since that deployment, the company has applied invisible watermarks to more than 180 billion images and videos, along with audio content equivalent to 240,000 years. Alphabet Chief Executive Officer Sundar Pichai announced at the Google I/O developer conference in May that the technology would be integrated natively into Google Search and the Chrome browser to let web users verify synthetic media instantly. Across Google Search, the Gemini application, and Chrome, the company's automated verification systems now process more than one million verification requests per day.

Technical boundaries and detection limitations

The detector relies strictly on locating the proprietary SynthID watermark embedded in a file's data structure, rather than scanning for visual flaws, anatomical distortions, or statistical generation artifacts. Because the tool only checks for SynthID signatures, it cannot detect synthetic media produced by other AI models that do not use the standard. Watermarks can also be removed from files in certain instances. A positive match indicates that an asset was made or edited with Google AI, but the system cannot determine whether an entire file was artificially generated or whether AI tools were merely used for minor edits. In addition, the system displays a warning that the submitted file may have been altered after the watermark was applied. Outside of Google, competing platforms maintain separate verification systems; Meta operates a dedicated detector for its Muse Image models, while OpenAI maintains a verification page supporting both SynthID and C2PA standards alongside its recent text watermarking initiative in the European Union.

This article is part of a thread

Culture, media and the new divides

What changed since this entry:

  1. 21/11Costa draft proposes 3% digital levy on large tech firms to fund the 2028–34 budget
  2. 21/11Weimer says investors ready to commit up to €25bn for 75–80% European ownership of TikTok Europe
  3. 21/11Czech licence-fee plan would cut public-media funding by about 15%

Ask about this text

Sources and context

Organizations
Google, Apple Inc., Nvidia

Pollar is free and carries no ads in the text. Its members keep it going.

Support Pollar