As artificial intelligence becomes increasingly capable of creating realistic images, videos, audio and text, one question is becoming more urgent: how can people tell what is real and what has been generated by a machine?
The challenge has grown significantly in recent years as AI image generators have advanced to the point where synthetic content can closely resemble authentic photographs. In response, technology companies are developing new methods to improve transparency and help users identify AI-generated media.
One of the most prominent solutions is SynthID, an invisible watermarking technology developed by Google DeepMind that is designed to leave a digital signature inside AI-generated content from the moment it is created.
Unlike traditional labels or metadata that can easily be removed, SynthID embeds information directly into content itself, allowing it to remain detectable even after certain edits or transformations. The technology is increasingly being viewed as a key tool in the global effort to establish trust and accountability in the age of generative AI.
What Is SynthID?
SynthID is a digital watermarking and detection system created by Google DeepMind to identify content generated by artificial intelligence.
Rather than attaching visible markers or relying solely on metadata, the technology embeds a machine-readable watermark directly into AI-generated content. These watermarks are invisible to human eyes and ears but can be detected by specialised software designed to verify whether a piece of content originated from supported AI systems.
Today, SynthID supports four major content categories: images, video, audio and text. It is used across Google's AI ecosystem, including Gemini, Imagen, Veo and Lyria. In 2025, Google announced that more than 10 billion pieces of AI-generated content had already been watermarked using the technology.
How SynthID Works for Images
For images, SynthID operates through a sophisticated watermarking process that subtly modifies pixel patterns within an AI-generated image.
The changes are carefully designed to be imperceptible to the human eye. While the image appears unchanged, a hidden signal is embedded throughout its pixel structure.
A separate detection system can then analyse the image and determine whether the watermark is present. If detected, the system can indicate that the image was generated or modified using an AI tool that supports SynthID.
 |
| Screenshot of Gemini’s response regarding an image. Image source: TechCabal |
The SynthID Detector
To make verification more accessible, Google introduced the SynthID Detector, a portal that allows users to upload content and check whether a SynthID watermark is present.
The system can analyse images, videos and audio files and identify regions where watermark signals are most likely embedded. This capability is particularly useful for journalists, researchers and fact-checkers attempting to verify the origins of digital content.
Google has positioned the detector as part of a broader effort to improve transparency across the rapidly evolving generative AI landscape.
Beyond Images: Watermarking Text, Video and Audio
Applying watermarking to text requires a different approach because text contains no pixels.
When an AI language model generates text, it selects words from a range of possible options. SynthID subtly influences these choices according to a hidden pattern that can later be identified by detection systems. The resulting text remains natural and readable to humans while carrying an invisible statistical signature.
For video and audio, similar principles apply, although the watermarking techniques differ according to the structure of each medium.
Google has also made parts of its text watermarking technology available to developers, encouraging broader adoption across the AI ecosystem.
OpenAI's Adoption of SynthID
A major milestone in SynthID's development came in May 2026 when OpenAI announced it would begin embedding Google's SynthID watermarks into images generated through ChatGPT, Codex and its API services.
The partnership marked one of the most significant collaborations in the AI industry and signalled growing support for shared content provenance standards. Alongside SynthID, OpenAI also adopted C2PA Content Credentials, creating a dual-layer verification system intended to help users trace the origins of AI-generated images.
According to OpenAI, SynthID provides a more durable verification signal because it remains embedded within the image itself, while C2PA metadata offers additional contextual information about how the image was created.
The move has fuelled speculation that SynthID could evolve from a Google technology into a broader industry standard for identifying AI-generated media.
What SynthID Cannot Do
Despite its growing importance, SynthID is not a universal AI detector.
The system can only identify content that has been generated using tools that intentionally embed SynthID watermarks. Images created by AI models that do not support the technology remain outside its scope.
The watermark is also not indestructible. Heavy editing, aggressive image processing, repeated transformations and deliberate attempts to remove or disrupt the watermark can weaken detection capabilities. Researchers and developers continue to test the limits of the technology, highlighting the ongoing arms race between content authentication systems and efforts to evade them.
Google has consistently emphasised that SynthID should not be viewed as a perfect solution but rather as one component within a broader ecosystem of tools designed to improve content authenticity and provenance.
Why SynthID Matters
Traditional AI detection systems often rely on identifying visual flaws such as distorted hands, unusual lighting or unnatural backgrounds. As AI models become more advanced, these indicators are becoming increasingly unreliable.
SynthID takes a fundamentally different approach. Instead of attempting to identify AI-generated content after it has been created, it embeds information about the content's origin directly into the generation process itself.
This shift from detection to provenance may prove crucial as AI-generated media becomes more common across journalism, entertainment, advertising and social media.
By helping users verify where content comes from, watermarking technologies such as SynthID aim to strengthen transparency, reduce misinformation risks and build trust in digital information.
As generative AI continues to evolve, so too will the technologies designed to authenticate it. What SynthID ultimately represents is a growing recognition across the technology industry that creators of AI systems have a responsibility not only to generate content, but also to help identify it.