What Is SynthID? How the SynthID Detector Works

SynthID is Google DeepMind's invisible watermark for AI-generated images, video, audio and text, added as the content is made so software can recognize it later. To check a file, use Google's SynthID Detector at synthid.com or ask the Gemini app whether Google AI made it. ScrubPhoto does not detect SynthID; it reads and removes a separate layer, the file's metadata, like C2PA Content Credentials.

Check and scrub your file here

It runs in your browser. Nothing is uploaded, and the result is lossless.

Choose photos or videos

Drop files here or paste an image. Photo on your phone? Open scrubphoto.com there.

No upload. Works offline once the page loads.
Lossless. Every pixel kept, byte for byte.
Free. No sign-up, no limits.

No photo handy?

What is SynthID?

Google DeepMind's SynthID page describes it as a way to watermark and later identify content made with AI. The mark goes in at the moment of creation. The FAQ on Google's detector site says it is woven into the pixels of images and video and into the frequency components of audio, so it moves with the content wherever the file goes.

It started small: DeepMind's August 2023 post released it as a beta for a limited number of Vertex AI customers using Imagen. By October 7, 2026, Google said it had watermarked more than 180 billion images and videos, plus 240,000 years of audio.

Which apps add SynthID watermarks

Start with Google, whose Gemini help center says every piece of AI media made or edited in Gemini Apps gets an invisible SynthID watermark. DeepMind's model pages say videos made with Veo are marked with SynthID and every Lyria music track is watermarked. The SynthID page adds audio from the podcast feature of NotebookLM and text from the Gemini app.

Other companies use it too, and the detector site names NVIDIA, OpenAI and Kakao. OpenAI's help center lists both SynthID and Content Credentials for images made in ChatGPT, Codex or its API, while noting that coverage varies by product, model and export path.

Apple is next, with a caveat. In June 2026, Apple announced that Apple Intelligence photo edits and Image Playground images would get a hidden SynthID watermark. Its September 2026 release note says SynthID will arrive in a software update later this year and cover most edited images, depending on the edits applied. Google's October post still lists Apple as coming soon.

How to use the SynthID Detector

The SynthID Detector is Google's own checker, at synthid.com. Google gave it to early testers in May 2025 and opened it to everyone on October 7, 2026, worldwide and in English.

Open synthid.com. On a first visit it shows Google's terms and a privacy notice, with an Agree and continue button.
Drag a file onto the page or tap Select file. The site lists images (JPG, PNG, WebP, HEIC, HEIF, AVIF, TIFF, BMP, GIF), video (MP4, MOV, WebM) and audio (WAV, MP3, OGG, FLAC, AAC, M4A).
Read the verdict. For an image it reports whether a watermark is present. For video and audio it points to the segments that carry one.

Read the fine print first. The privacy notice mentions signing in and daily usage quotas, and says uploaded files are deleted once results come back. The extra terms forbid using the detector to probe for ways to break the watermark. Google also asks for the highest quality file you have, because heavy compression and editing affect detection.

How to check for SynthID in the Gemini app

You can also ask Gemini, and Google's verification help page gives these steps on a computer. Go to gemini.google.com and click Add files. Upload from your device, Google Photos or Google Drive. Then ask something like "Was this image created or edited by Google AI?" or "Is this AI-generated?"

You must be signed in with a personal account or a qualified Workspace account. Gemini checks one file at a time, up to 100 MB, with video under 90 seconds and audio under an hour, and allows about 10 checks per media type in a rolling 24 hours.

Google says the portal joins checks already built into Search, the Gemini app and Chrome. For a file you think came from ChatGPT, OpenAI points to its own tool at openai.com/verify, which looks for OpenAI's SynthID watermark and its C2PA manifest.

What a SynthID result means, and what it does not

A SynthID checker answers one narrow question, which is whether a SynthID watermark is in this file, and Google's pages explain each possible answer.

Watermark detected. All or part of the content was likely created or edited by a model that uses SynthID. For video, Gemini says which parts.
Not detected. No confident signal was found. The file can still be AI-made, by a model that never used SynthID, before a company adopted it, or after heavy changes weakened the signal.
Unclear. Gemini gives reasons like too little detail to hold a watermark, or an edit too small to mark.

A hit tells you a tool was involved, not who used it. OpenAI says its signals hold no details on who made a file or what they typed, and that a watermark does not show how much of a file the model made. Google's detector FAQ recommends gathering more evidence before you decide anything.

SynthID vs C2PA metadata

Many AI files carry two marks at once. Google's help says Gemini Apps add C2PA Content Credentials as well as SynthID, and OpenAI says the same of ChatGPT images. Content Credentials are signed metadata stored beside the image data. SynthID lives inside the image data itself.

That difference decides what survives. OpenAI's help center warns that a screenshot or a format conversion can strip Content Credentials, while a watermark may outlast some changes that drop metadata. Our guide to what C2PA Content Credentials are covers the metadata side.

Platforms also read different layers. Meta says the AI info label on Instagram (Instagram has also shown it as AI content, and first called it Made with AI) comes from C2PA and IPTC metadata in the file or from the person posting. The Instagram page explains how that plays out. Platform rules can also require you to disclose realistic AI video or audio, whatever the file says.

What ScrubPhoto checks, and what it leaves to Google

ScrubPhoto is a metadata tool, not a SynthID checker. The checker opens a file on your device, with no upload, and lists its metadata: any C2PA record and the app that signed it, the IPTC field that says whether AI made or edited the image, the XMP creator tool, and EXIF camera details.

It never looks for SynthID and cannot remove it. A JPEG, PNG or WebP is scrubbed without decoding the image, so the copy you download matches the original pixel for pixel, and any watermark in those pixels stays put. A clean metadata readout does not mean a post will go unlabeled.

To see what a Gemini export carries, read the Gemini page. The limits page covers everything outside a metadata tool's reach. If you came here to get rid of the watermark, can you remove SynthID gives the straight answer.

Questions about SynthID

Can you see a SynthID watermark?

No. Google says the mark is imperceptible to people, but SynthID's own technology can detect it. Some apps add a separate visible logo, and Gemini's help page says the setting for that visible mark does not affect SynthID.

Does ChatGPT use SynthID?

Yes, for images and audio. OpenAI's help center says images made with ChatGPT, Codex or the OpenAI API include SynthID alongside Content Credentials, and ChatGPT audio includes a SynthID watermark.

Who can use the SynthID Detector?

Since October 7, 2026, anyone. Google opened synthid.com worldwide in English after a 2025 rollout to early testers, with a waitlist for journalists, media professionals and researchers.

Can the SynthID Detector be wrong?

Yes, in both directions. A missing watermark does not rule out AI, and Google's FAQ says that in very rare cases SynthID can falsely trigger on content with no watermark. Treat any result as one piece of evidence.

Is the SynthID Detector a general AI detector?

No. It finds only media from companies that have adopted SynthID, which Google lists as itself, NVIDIA, OpenAI and Kakao, with Apple coming. Google's FAQ explains the difference: general detectors estimate from patterns a model leaves behind, while SynthID looks for a signal planted on purpose.

Can ScrubPhoto check a file for SynthID?

No. It reads and removes metadata only. To check for SynthID, use Google's SynthID Detector or the Gemini app, or openai.com/verify for files from OpenAI.

Sources

Official pages this page was checked against in October 2026: