Guide · Updated 2 November 2026

SynthID explained

SynthID is the invisible watermark Google puts in content made by its AI, and that other companies now use too. Here's how it works and how to check for it.

Short answer

SynthID is an invisible watermark developed by Google DeepMind. It is woven into AI-generated images, video, audio and text so that software can later recognise them, even after cropping, filters or compression. Anyone can check an image, video or audio file for it with Google's free SynthID Detector at synthid.com.

How SynthID works

When a SynthID-enabled tool makes an image, it adjusts the pixels in a pattern that's invisible to people but recognisable to a matching detector. For audio, the pattern sits in the sound; for video, in each frame. For text, it gently shifts the generator's word choices so the statistics carry a signal.

Because the mark is part of the content, not the file around it, it survives the things that strip metadata: screenshots, re-saving, uploading to social media and moderate editing.

Who uses it

Google applies SynthID to content made by its own AI models, including Gemini's image and video tools. Other companies have adopted it for their own output, including OpenAI for images from ChatGPT and its API, NVIDIA and Kakao, and Google has named Apple as an upcoming partner. By October 2026 Google reported more than 180 billion images and videos and 240,000 years of audio watermarked.

How to check for it

  • SynthID Detector: upload an image, video or audio file at synthid.com. Google opened it to everyone worldwide on 7 October 2026, in English, with a daily limit on checks.
  • Gemini app: upload an image and ask whether it was made with Google AI. This has worked since November 2025.
  • Search and Chrome: Google has been adding SynthID checks to its search results and browser.

Reading a result

Result What it means
SynthID detected Made or edited with a tool that uses SynthID
Partly detected Some regions or segments carry the mark, for example an edited photo
Not detected No SynthID found. It may still be AI-made by another tool, or edited heavily

SynthID and Content Credentials together

The two systems cover each other's weaknesses. Content Credentials carry rich detail (which tool, when, what edits) but disappear when metadata is stripped. SynthID carries a simple signal but survives. For the full picture, check both: the file's own labels with our AI image checker, then the pixels at synthid.com.

SynthID for text

Google has watermarked text from the Gemini app with SynthID since 2024, and has released SynthID Text as open source, for example in Hugging Face Transformers. Google's own documentation is frank about the limits: it survives small edits and light paraphrasing, but thorough rewriting or translation weakens it a lot. For what this means for ChatGPT and other chatbots, see does ChatGPT watermark text?

Questions

Can you see a SynthID watermark?

No. It's built into the pixels, sound or word choices in a way people can't perceive. Only Google's detector can read it.

Can SynthID be removed?

It's designed to survive common edits like cropping, colour changes, filters, screenshots and compression. Very heavy changes can weaken it. We don't offer removal, and we don't think anyone should try to hide that content was AI-made.

Does no SynthID mean an image is real?

No. SynthID only marks content from tools that use it. An image from another generator, or one edited heavily enough, won't show a result.

Is SynthID the same as Content Credentials?

No. Content Credentials (C2PA) are a signed record in the file's metadata, readable by anyone but easy to lose. SynthID is a watermark in the content itself, harder to remove but only readable by Google's detector. Many images now carry both.