How to Check If an Image Is AI-Generated: Google’s SynthID Detector Explained

AI-generated images are now realistic enough to fool anyone at a glance. Google’s answer is SynthID — an invisible watermark embedded in content made by its AI models — and a free public SynthID Detector that anyone can use to check a suspicious file. Here’s how to use it, where else verification shows up, and what the tool can’t tell you.

What Is SynthID?

SynthID is Google DeepMind’s watermarking technology. It embeds an invisible signal directly into AI-generated content — between the pixels of an image, in the waveform of audio, in frames of video — that survives common edits like cropping or color changes. Google says more than 180 billion watermarked assets have been processed to date.

The watermarks now come from Google plus partners including OpenAI, NVIDIA and Kakao, with Apple expected to join. That means the detector can flag content from several major AI systems — but far from all of them.

Method 1: Use the SynthID Detector Website (Easiest)

1. Open synthid.com in your browser (free, available worldwide, currently English-only).
2. Sign in with a Google, OpenAI or Apple account.
3. Upload the image, video or audio file you want to check.
4. Read the result: “watermark detected” means a SynthID mark from Google or a partner was found; “watermark not detected” means no compatible mark exists in the file.

Method 2: Ask Gemini Directly

1. Open the Gemini app or website.
2. Tap Add files and upload the suspected image.
3. Ask a verification question such as “Was this created with Google AI?”, “Is this AI-generated?” or type @synthid after attaching the image.
4. Gemini scans for the SynthID watermark and uses its own reasoning to answer — it can even estimate what percentage of the image was AI-created or edited.

Google says this verification feature inside Gemini had already been used 50 million times as of May 2026.

Method 3: Check in Search, Chrome and YouTube

SynthID checks are spreading across Google’s products:

• Google Search / Lens / Circle to Search: right-click an image or use Circle to Search to check for SynthID or C2PA markers.
• Chrome: verification built into the browser via Gemini.
• YouTube: since 2026, YouTube automatically detects and labels videos containing significant photorealistic AI content using SynthID and C2PA signals.

What “Watermark Not Detected” Really Means

This is the single most misunderstood part. A negative result is not proof the file is human-made. It only proves there is no SynthID watermark in it. An image from any generator that doesn’t watermark — or a genuine photograph — returns the same “not detected” result.

Think of it as a one-way test: positive = almost certainly AI-generated (by a watermarking system); negative = inconclusive. Google itself describes SynthID as one building block for AI transparency, not a complete solution.

Limitations You Should Know

• Not universal: only detects watermarks from systems that embed SynthID. Content from unwatermarked models slips through.
• Can be degraded: heavy editing, severe cropping, pitch-shifting audio or paraphrasing text can weaken or strip the signal. Researchers have demonstrated attack tools that sharply reduce detection rates.
• Complementary standard: Google also supports C2PA content credentials, which add provenance metadata (e.g. “shot on a Pixel camera, then AI-edited”). The two systems work best together.

For broader protection, pair verification habits with general online safety practices — skepticism about viral media is still your best first filter.

Frequently Asked Questions

Is the SynthID Detector free?

Yes. The public detector at synthid.com is free to use worldwide; you just sign in with a Google, OpenAI or Apple account.

What file types can it check?

Images, video and audio. Text watermarking exists as an open-source detector for developers, but the public tool focuses on media files.

Can SynthID detect ChatGPT-generated images?

OpenAI has committed to adopting SynthID starting with images created through ChatGPT, Codex or the OpenAI API, so those should be detectable.

If no watermark is found, is the image definitely real?

No. “Not detected” only means no SynthID watermark was present — the file could still be AI-generated by a system that doesn’t watermark, or it could be genuine.

Does it work on videos and audio too?

Yes — the public detector accepts video and audio uploads in addition to images.

Sources

• TechTimes — Google SynthID AI-Detector Expands Globally
• 9to5Google — How to use Google Gemini’s built-in AI image detection tool
• DevDossier — Google SynthID Explained (2026)