What Is SynthID? Google’s AI Watermark Explained

Guide · Updated October 8, 2026
SynthID is a watermarking technology from Google DeepMind. It embeds a signal that people cannot see or hear into images, video, audio and text produced by AI models, so the content can later be identified as AI-generated by a matching detector.
Images and video. A neural network adjusts pixel values by amounts too small to notice. The pattern is spread across the whole frame, so it survives compression, resizing, cropping, colour filters and screenshots far better than metadata does. A second network trained alongside it reads the pattern back and reports whether it is present.
Audio. The signal is embedded in the waveform in a way that is inaudible and designed to survive common edits such as added noise, MP3 compression and speed changes.
Text. SynthID Text slightly adjusts the probabilities a language model uses to pick each word, leaving a statistical pattern without changing the meaning. Detecting it requires the watermarking keys of the model that wrote the text.
Google applies SynthID across its generative products, including the Gemini app (Nano Banana image generation), Imagen and Veo on Vertex AI, and Lyria music. Since May 2026 OpenAI embeds it in images from ChatGPT, Codex and the OpenAI API, and other partners have adopted it for their media. Google reports that well over a hundred billion pieces of content carry the watermark.
Google and OpenAI do not rely on the watermark alone. Their images also carry C2PA Content Credentials — a signed manifest that says the image was created by generative AI. Google’s manifest literally records “Applied imperceptible SynthID watermark”. Unlike the watermark, the manifest is removed by screenshots and most social platforms, but while it is there it is cryptographically verifiable and names the signer.
1. Upload the original file to the SynthID Detector on this site. If the Content Credentials are intact you will see whether a SynthID watermark was declared, who signed it and whether the signature is trusted.
2. If no credentials remain, upload the same file to Google’s official SynthID Detector or ask the Gemini app “Was this created with Google AI?”. Google answers watermarked, not watermarked or uncertain.
3. For files from tools that do not use SynthID, an AI visual analysis can point out generation artifacts, but treat it as an estimate.
SynthID was developed by Google DeepMind. Google applies it to content from its generative AI products, and partner companies — OpenAI among them since 2026 — apply it to their own AI output.
No. It is designed to be imperceptible to people. It is spread across the pixels of an image or video frame, or the waveform of audio, and only a matching detector can recover it.
It is built to survive common changes such as compression, cropping, resizing, filters and screenshots. Heavy edits can weaken it, which is why Google’s detector can answer “uncertain”. No watermark is impossible to defeat.
No. SynthID is a watermark in the content itself. Content Credentials (C2PA) are a signed manifest stored next to the content. Google adds both, and its Content Credentials state that a SynthID watermark was applied — the record the SynthID Detector on this site reads.
No. It only identifies content from companies that embed SynthID. An image from a model that does not use it will never show a SynthID watermark, even though it is AI-generated.
SynthID Detector is an independent service and is not affiliated with Google, Google DeepMind or OpenAI. SynthID is a Google technology. Results describe the provenance data found in a file and are not legal proof of how an image was made.
SynthID Detector is an independent service and is not affiliated with Google, Google DeepMind or OpenAI. SynthID is a Google technology. Results describe the provenance data found in a file and are not legal proof of how an image was made.