Google DeepMind announced the global availability of SynthID Detector, a tool designed to help identify AI-generated content, including media created using AI tools from SynthID partners such as OpenAI, NVIDIA, and Kakao, with Apple support expected soon.
Before the introduction of SynthID Detector, users could identify content AI-generated only by asking Gemini, which was limited to content generated by Google AI. Identifying content from other AI providers required instead using their respective detection tools.
The SynthID.com website simplifies the process by allowing users to upload an image, audio, or video files to have it checked across all SynthID partners, including OpenAI, NVIDIA, Kakao, and (soon) Apple. The tool provides a yes or no outcome, albeit Google also points out that when "no confident signal" is found, a negative outcome does not necessarily mean the content is not AI-generated:
This is not a general AI detector. We can only detect media from the above companies, who have adopted SynthID technology. We are continuously working on extending our partnerships to allow detecting more generated media.
Namely, SynthID does not cover content generated using Meta's or Microsoft's AI, which uses a different watermarking technology. In addition, extensive modifications can also degrade SynthID's watermark signal, resulting in false negatives. In such cases, users need to apply a more comprehensive approach involving additional tools. Despite these limitations, SynthID was included in TIME's Best Inventions of 2026 list, which recognized its approach to AI-content detection over AI detectors.
Commenting on the announcement, Reddit user Chicano_Ducky expressed a negative sentiment about the efficacy of this approach:
[T]hey never worked in the first place, the false positives are unacceptably high on videos their own models made which is in the article [...] They are trying to put the genie back in the bottle and they can't.
User Bulky-Employer adds yet another dimension to this discussion emphasizing that:
A removal node is easily created. You just run a low denoising pass on the image. You can already test this on OpenAI and Gemini's outputs.
SynthID is just a speed bump against people who want to defraud using AI images. It's a good thing that it exists.
A complementary approach to watermarking AI-generated content is C2PA metadata, which records information about a file’s origin and editing history to help establish its authenticity. Harry G argues on LinkedIn that this is the only "approach that actually scales", as "trying to detect synthetic content after the fact turns into an arms race". However, C2PA metadata however can be lost when content is uploaded and does not carry over to screenshots.
Launched in 2023 by Google DeepMind, SynthID is a cryptographic watermarking and detection solution for AI-generated images, audio, video, and text. It embeds an invisible watermark directly into the content by altering audio frequency probability distributions or pixel/token patterns using a secret key.
About the Author
Sergio De Simone
Show moreShow less