SynthID Detector
Back to guides

Was This Image Made with Gemini? Check It with the SynthID Detector

Nano Banana and Google's other image models hide a SynthID watermark in every picture. How to check an image for it, step by step, and how to read the answer.

Oct 21, 2026

If you want to know whether a picture came from Gemini, the most reliable check is Google's own SynthID Detector. Images made with Gemini's image models, including Nano Banana, Nano Banana Pro and Nano Banana 2, carry an invisible SynthID watermark that Google can read. This guide shows what to check first, how to run the watermark check and how far you can trust the result.

The short answer

  1. Drop the original file into the free metadata checker on our homepage to see whether it still declares Google AI.
  2. Upload the same file to Google's SynthID Detector at synthid.com, or ask the Gemini app "Was this made with Google AI?"
  3. Treat "detected" as strong evidence of a SynthID-enabled model, and "not detected" as no evidence either way.

What Gemini leaves inside an image

A Gemini image can carry three kinds of evidence. They fail in different ways, so it pays to check more than one.

  • The SynthID watermark. Google DeepMind's SynthID hides an imperceptible signal in the pixels. It is designed to survive compression, resizing, cropping and filters, but only Google's tools can read it. Google has not released a public detection API, so no third-party site, including this one, can read the watermark directly.
  • Content Credentials (C2PA). Signed metadata that can name the company behind the file. Precise, but screenshots and most social uploads strip it.
  • AI labels. Plain metadata fields such as "Made with Google AI", or a generator tag that names Nano Banana or Imagen. Easy to read and easy to remove.

Step 1: read the metadata first

Open the checker on our homepage and drop in the original image. It runs in your browser, so the file is not uploaded anywhere. If the metadata survived, the report may show a Google signature in the Content Credentials, a digital source type of "Created by a trained AI model", a "Made with Google AI" label, or a Nano Banana or Imagen tag.

If the report says no provenance data was found, don't read anything into it yet. Most platforms strip metadata on upload, and a screenshot never carries it. The watermark lives in the pixels themselves and can survive where the metadata did not.

Step 2: check the image with Google's SynthID Detector

  1. Go to synthid.com and sign in with a Google, Apple or OpenAI account. The service is free and works worldwide in English, with a daily limit on checks.
  2. Upload the best copy you have. JPG, PNG, WebP, HEIC, AVIF, TIFF, GIF and BMP files up to 100 MB are accepted. Use the original download rather than a screenshot or a re-shared copy, because heavy compression and editing weaken the watermark.
  3. Read the answer. Images get a "detected" or "not detected" result.

Google says it processes the file in real time, deletes it after returning the result and keeps a signature of it for 24 hours to enforce the quota.

Prefer the Gemini app?

The Gemini app runs the same check for Google AI content. Upload the image and ask "Was this made with Google AI?" It allows about 10 checks of each content type per 24 hours. It only covers Google's own models, so use synthid.com when the image might come from somewhere else, such as ChatGPT.

How to read the SynthID Detector result

  • Detected. A SynthID mark is present, so the image was likely generated or edited by a SynthID-enabled model. That is not only Gemini: images from OpenAI carry SynthID too, so look at the Content Credentials when you need to know which company was involved. The result also does not say how much was changed, which means a real photo edited with a Gemini image model can come back as detected.
  • Not detected. No confident signal. The picture may be human-made, made by a model that does not use SynthID (Midjourney, Stable Diffusion, FLUX, Grok and most open models), or edited heavily enough to weaken the mark. It is not proof that the image is real.
  • A rare false positive. Google notes that in very rare cases SynthID can trigger on content without a watermark. Check the source and the context before you act on a single result.

Common questions

Does cropping or a screenshot remove the Gemini watermark?

SynthID is built to survive common changes like compression, resizing, cropping and filters, though many stacked edits can defeat it. A screenshot almost always strips the Content Credentials and AI labels, which is why the metadata check and the watermark check complement each other.

Can I check a Gemini image without signing in?

The metadata checker on our homepage needs no account and has no limit. Reading the watermark itself takes Google's tools, and both synthid.com and the Gemini app need a sign-in.

Where can I learn more?

The SynthID Detector FAQ covers privacy, AI text and other generators, and the supported models list shows every company whose content carries SynthID. Checking a picture from OpenAI instead? Read Do ChatGPT images have SynthID?

SynthID Detector is an independent site, not affiliated with Google. SynthID is a trademark of Google LLC. Limits are as Google published them in October 2026.