SynthID Goes Industry-Wide—OpenAI and Nvidia Just Signed On
Google's SynthID watermark has labeled 100B AI images. Now OpenAI, Nvidia, Kakao, and ElevenLabs are adopting it—and a new detector portal opens to the public.
In Brief
- Google says SynthID has watermarked 100 billion AI-generated images and videos, plus 60,000 years of audio since 2023
- OpenAI will embed SynthID in GPT 2 images, Nvidia in Cosmos world models, and Kakao and ElevenLabs in their respective AI outputs
- Google launched a SynthID Detector portal and is integrating watermark scanning into Circle to Search, Lens, and Chrome—but there is no public API, and open models remain unmarked
Google’s SynthID watermark has been applied to 100 billion AI-generated images and videos plus 60,000 years of audio since its 2023 debut. Now the technology is expanding beyond Google’s own products. OpenAI will add SynthID to images produced by GPT 2, Nvidia will embed it in Cosmos world foundation model outputs, and Kakao and ElevenLabs will watermark their AI content as well.
SynthID embeds a digital watermark directly into the pixels of images, the waveforms of audio, and the frames of video—meaning it survives compression, cropping, and rotation. Google DeepMind scientist Pushmeet Kohli says the team built SynthID to resist attacks. Google says no successful bypass has been demonstrated, despite claims from a few tinkerers.
Detection Expands—But Gaps Remain
Google also launched the SynthID Detector portal, where anyone can upload content to check for the watermark. Scanning is being integrated into Circle to Search, Google Lens, and AI Mode. Later this year, Chrome and Search will gain C2PA-based provenance checks that read metadata describing how content was created—including AI involvement.
Nvidia was the first external company to adopt SynthID, adding it to Cosmos video outputs in March 2025. The difference now is scale: with OpenAI on board, SynthID covers the two largest AI image generators on the internet. The catch is that open-source models still produce AI content with zero watermarking, and most AI content currently online came from tools outside any watermarking ecosystem.
The dual-track approach—SynthID’s embedded watermark plus C2PA’s visible metadata—reflects a recognition that no single method is sufficient. The EU’s AI Act enforcement delays have given companies breathing room on labeling mandates, but voluntary adoption may end up shaping de facto compliance. As arXiv’s year-long ban for AI-slop submissions shows, the pressure to distinguish human from machine output is intensifying across every domain.
FAQ
What is SynthID?
A digital watermarking system developed by Google DeepMind that embeds invisible markers into AI-generated images, video, and audio. It survives compression, cropping, and rotation.
Which companies are adopting SynthID?
OpenAI (GPT 2 images), Nvidia (Cosmos world foundation models), Kakao, and ElevenLabs. Google has applied SynthID across its own AI products since 2023.
Can I check if an image has a SynthID watermark?
Yes. Google launched the SynthID Detector portal for public uploads. Scanning is also being added to Circle to Search, Google Lens, and AI Mode, with C2PA checks coming to Chrome and Search.
Does SynthID cover all AI-generated content?
No. Only content from participating platforms is watermarked. Open-source models and tools outside the SynthID ecosystem produce AI content without any watermark.
