Updated
Updated · Ars Technica · Jul 29
Google Expands SynthID to Label 100 Billion AI Images and Videos via New Partnerships
Updated
Updated · Ars Technica · Jul 29

Google Expands SynthID to Label 100 Billion AI Images and Videos via New Partnerships

3 articles · Updated · Ars Technica · Jul 29

Summary

  • Google is widening use of its SynthID watermarking system through partnerships with OpenAI, Runway, Nvidia and others to label AI-generated media across more tools.
  • 100 billion AI images and videos have already been created with Google tools in just a couple of years, underscoring the scale the company says makes reliable labeling urgent.
  • SynthID embeds invisible markers in pixels, video and audio waveforms, and Google says the watermark can survive compression, resizing and edits that often strip other labels.
  • C2PA metadata remains part of Google’s labeling approach, but it can disappear after simple edits or screenshots, leaving SynthID’s durability as the key test for broader industry adoption.

Insights

With AI generating billions of images, will embedded pixels or human intuition be our final defense against deception?
If invisible watermarks are the ultimate defense against AI fakes, what happens when hackers inevitably break the code?