Nano Banana Detector

Add the picture to the Nano Banana detector above. Your browser looks inside the file for Content Credentials signed by Google LLC. If they are there, say the image was AI-generated and still match the file, you get Likely AI-generated straight away, with no detection model and no free check used. Without them, a detection model judges the pixels.

Nano Banana is Google's name for the image models in Gemini. We tested three: Nano Banana (Gemini 2.5 Flash Image), Nano Banana 2 (Gemini 3.1 Flash Image) and Nano Banana Pro (Gemini 3 Pro Image). The credentials step runs in your browser, so the image isn't uploaded for it. The model step uses Sightengine, with Hive as a backup, the same engines as the AI image checker on our home page.

In our test, 6 of the 8 original Nano Banana files carried Google-signed credentials, and the checker settled those 6 without a model. The other 4 images, 2 dataset copies and 2 screenshots, went to Sightengine, which called every one AI. Its lowest score on any of the 10 was 0.99. The checker got 10 right out of 10, and Hive also caught all 10.

The split by source is the interesting part. All 6 images we generated ourselves through Google's API (via a reseller that relays requests to it) carried credentials signed by Google LLC, naming Google C2PA Core Generator Library as the tool, and Google is on the current C2PA trust list. The 2 copies we took from a public dataset of images from the first Nano Banana model, posted in August 2025, carried none. We can't say whether those files lost their credentials on the way into the dataset or never had them. Google said in November 2025 that it was starting to embed C2PA metadata in images from Nano Banana Pro in the Gemini app, Vertex AI and Google Ads.

Google also hides an invisible SynthID watermark in images its AI makes. Checking it takes Google's own tools: you can upload an image to the Gemini app and ask whether it was made with Google AI, and Google's SynthID Detector does the same job. Our checker doesn't read SynthID at all, and we don't claim to.

That matters most for copies. Google says the watermark usually survives resizing, color changes and compression, though after many changes it may no longer be detected, so a screenshot that has lost its credentials may still carry it. If you suspect a picture came from Gemini, the Gemini app can answer about Google AI specifically. Two limits apply: Google says the Gemini app currently recognizes only content made with Google AI tools, and a missing watermark doesn't make a picture real.

Use the check that fits the copy you have. The Nano Banana detector above works on any image from any generator; Google's tools answer one narrower question, whether Google AI made or edited the picture.

If you mostly see images from OpenAI's tools, the ChatGPT image page explains OpenAI's signals and its own verification tool.

A screenshot strips the credentials, so the Nano Banana detector falls back on its model. Neither of our 2 Nano Banana screenshots kept Google's credentials, and Sightengine still called both AI, as it did every Nano Banana image we gave it. Hive did the same.

Two screenshots is a tiny sample, so don't read it as a promise. Screenshots shrink the picture, add a status bar and often get compressed again when shared, and any of that can weaken the traces a detector looks for. Our guide to viral pictures and screenshots covers how to find a better copy.

A valid Google credential that says AI-generated tells you the file came out of Google's image service and hasn't changed since it was signed. It doesn't tell you who asked for it, what they asked, or whether the scene shows something that happened.

If someone edits the file after signing, the credentials no longer match the image, and the checker stops relying on them and asks the model instead. A picture in which the Nano Banana detector finds no Google credentials isn't cleared either: in our test, 2 of the 8 original files had none. To see how another company handles signing, read about FLUX images, or read a file's credentials yourself.

No. Reading SynthID on a Gemini image takes Google's tools, such as the Gemini app or Google's SynthID Detector. Our checker reads Content Credentials and metadata in your browser, then asks a detection model if those don't settle it.

The 2 originals without them came from a public dataset. Files can lose credentials when they are re-saved or converted, and older Nano Banana images may never have had them. Every image we generated through Google's API had them.

If the credentials say AI-generated, the signature is valid and the file hasn't changed, yes, without a model. If they say the picture was edited with AI, you get Edited with AI. If the file changed after signing, the model decides.

No. Credentials are read in your browser. If the image needs a model, our server passes it once to Sightengine (or Hive as a backup) for a score and doesn't save it. See the privacy page.