What Is SynthID? How to Use SynthID Detector to Check for Google's AI Watermark, and What a Result Means

Last reviewed October 19, 2026 · 7 min read · By the PEA team (Lumina Spark)

日本語版 (Japanese version)

Short answer: SynthID is an invisible watermark from Google DeepMind that's embedded in images, video, audio and text made with Google's AI. Since October 7, 2026, anyone can use SynthID Detector (synthid.com, in English) to check whether an image, video or audio file was made with AI from Google or its partners, including OpenAI, NVIDIA and Kakao, with Apple to follow. The Gemini app can check too, but only for Google's AI. Neither is a general AI detector: if no watermark is found, that doesn't mean the content is real.

What SynthID is

Google DeepMind describes SynthID as digital watermarks embedded directly into AI-generated images, audio, text and video. People can't see or hear them, but SynthID's technology can detect them. For images and video, the watermark is added the moment the content is created and is designed to stand up to cropping, filters, frame-rate changes and lossy compression.

SynthID started in August 2023 with a limited number of Google Cloud (Vertex AI) customers using the Imagen image model. By October 2026, Google said it had watermarked over 180 billion images and videos and 240,000 years of audio.

Which AI tools add SynthID

CompanyWhat carries SynthIDSource
GoogleAll AI media created or edited in Gemini Apps, and Google's other generative AI productsGoogle
OpenAIImages made with ChatGPT, Codex or the OpenAI API (from May 2026); supported audio (from July 31, 2026)OpenAI
NVIDIAOutput of its Cosmos world foundation models; announced in March 2025 as the first external user of SynthIDNVIDIA
Kakao (South Korea)Its image and video models, from the second half of 2026 (announced May 2026)Kakao (Korean)
AppleImages made with Image Playground and photos edited with Apple Intelligence in Photos (announced June 2026); Google said on October 7 that detection is coming soonApple, Google

Content from AI tools that haven't adopted SynthID carries no SynthID watermark.

How to check for SynthID

1. SynthID Detector (synthid.com)

  • What it checks: images, video and audio made with AI from Google and the companies above. It opened to everyone, worldwide and in English, on October 7, 2026.
  • How: upload a file at synthid.com. On first use, you're asked to agree to the terms. The site recommends uploading the highest-quality file you have.
  • Results: for an image, "Watermark detected" or "Watermark not detected." For video and audio, the segments that carry the watermark.
  • Your file: the site's privacy notice says uploaded files are deleted as soon as the result is returned. A digital signature of the file is kept for 24 hours to enforce daily usage quotas.
  • File types: the upload accepts common image formats (including JPEG, PNG, WebP and HEIC), video (MP4, MOV, WebM) and audio (including WAV and MP3), as of October 2026. It doesn't check text.
  • No bulk use: the terms forbid automated or bulk submissions and using multiple accounts to get around the quotas.

2. The Gemini app

  • What it checks: only content made with Google's AI. Google's help page says that while other companies have started to adopt SynthID, Gemini can currently only recognize content created by Google AI tools. If a file has C2PA Content Credentials, Gemini can also summarize them.
  • How: at gemini.google.com or in the app, add one file and ask something like "Is this AI-generated?" Image checks arrived in November 2025 and video checks in December 2025.
  • Limits (Gemini Apps Help): one file at a time, 100 MB or less. Videos must be under 90 seconds and audio under 1 hour. There's a quota of about 10 image checks, 10 video checks (up to 5 minutes in total) and 10 audio checks (up to 3 hours in total) in any rolling 24 hours.
  • Screenshots: crop tightly around the image, and don't upload a collage of several images.

Google says it also has built-in verification features in Search and Chrome, and that together with the Gemini app they handle over 1 million requests a day. OpenAI offers its own public verification tool (a preview) for images and supported audio made with its tools.

What a result tells you, and what it doesn't

What it can tell youWhat it can't, and caveats
"Watermark detected": the content was likely generated or edited by an AI model that uses SynthIDIt isn't a general AI detector. It only finds watermarks from Google and its partners
For video and audio: which segments carry the watermark"Not detected" doesn't rule out AI generation or manipulation. The content may come from a model without SynthID
Heavy edits and compression can weaken the watermark until it isn't detected. Someone who tries many transformations may find one that defeats it
In very rare cases, SynthID can falsely trigger on content without a watermark

The SynthID Detector page itself recommends gathering additional evidence. Don't accuse or punish anyone on a watermark result alone. For the limits of detectors in general, see How Accurate Are AI Image Detectors?

SynthID vs. C2PA Content Credentials

  • SynthID: an invisible watermark inside the content itself, designed to survive re-saving and common edits.
  • C2PA Content Credentials: a signed record of how the content was captured, generated or edited, attached to the file as metadata. Social networks often strip metadata on upload, and then there's nothing left to check.

Google uses both. Gemini Apps add SynthID and C2PA Content Credentials to the media they create, and Google says turning off the visible watermark in Gemini doesn't affect either. Pixel 10 phones add C2PA Content Credentials to photos taken with Pixel Camera. To check a photo for camera data and C2PA, see How to Check If a Photo Is AI-Generated Before Using It as Evidence.

SynthID for text

SynthID also watermarks AI-generated text. In October 2024, Google made SynthID Text available to all developers through Hugging Face and its Responsible GenAI Toolkit. As of October 2026, SynthID Detector doesn't check text.

How PEA approaches this

When PEA can confirm AI generation from a marker embedded in an image (a watermark or an AI-generation record), it bases its verdict on that marker. When there's no marker, it lists confirmed facts and concerns separately, based on C2PA signatures, camera data and eight independent analyses, and grades the image from 1 to 10. When there isn't enough to go on, it says "Not assessed" instead of guessing. Any method can miss things or be wrong, PEA included. The first check is free and in English, with no account: try a free check.

Frequently asked questions

What is SynthID?

An invisible digital watermark from Google DeepMind, embedded in AI-generated images, video, audio and text. People can't see it, but SynthID's technology can detect it. The image and video watermark is designed to survive common edits such as cropping, filters and lossy compression.

How do I check if an image has a SynthID watermark?

Upload it to SynthID Detector at synthid.com, which has been open to everyone since October 7, 2026. It checks images, video and audio made with AI from Google and its partners: OpenAI, NVIDIA and Kakao, with Apple to follow. You can also upload a file in the Gemini app and ask "Is this AI-generated?", but Gemini only recognizes content made with Google's AI.

If SynthID Detector finds no watermark, is the image real?

No. Google says a "not detected" result doesn't rule out AI generation or manipulation. The image may come from a model without SynthID, or heavy edits may have weakened the watermark. Check the original file, run a reverse image search, look for C2PA data, and ask where the image came from.

Can SynthID Detector check ChatGPT images?

Yes, for images made after OpenAI adopted SynthID. OpenAI announced in May 2026 that images generated through ChatGPT, Codex and the OpenAI API carry SynthID, and OpenAI is one of the partners SynthID Detector covers. Images made before that, or with tools that haven't adopted SynthID, have no SynthID watermark, and heavy edits can weaken it.

Is SynthID the same as C2PA?

No. SynthID is an invisible watermark inside the content itself, designed to survive re-saving and common edits. C2PA Content Credentials are a signed record attached as metadata, which social networks often strip. Google adds both to media created with Gemini Apps.

Sources