DeepFake Check

Nano Banana Detector

Check whether an image may have been generated or edited with Nano Banana or Gemini. Upload the file to the image detector. The check reads an embedded C2PA credential and visual signals. It does not decode SynthID.

Limited free allowance, no signup. The result is screening evidence, not proof a photo is real.

Check this image

What this check can actually read

Nano Banana is Google’s image model inside Gemini. People search for a nano banana detector, a nano banana image detector, or a Gemini AI image detector when a picture looks too polished to trust. DeepFakeCheck does not have a separate Google-only model. The same AI image detector is the tool: you upload a JPG, PNG, or WebP file up to 50 MB.

In the browser, that upload is read for an embedded C2PA Content Credential, also called a manifest. If the manifest is there, the reader reports whether the signature validates, who the claim generator and signer are, and which actions the credential records. A trusted credential that declares the whole file was created by generative AI is reported as such. A credential that records generative editing or compositing is not treated as “the entire picture was generated.” A missing credential is not evidence that the image is fake, and a valid credential is not proof that the scene happened.

Gemini and Nano Banana credentials

Google has said some Gemini and Nano Banana images include C2PA Content Credentials. This checker will read that manifest when it is still inside the file you upload, the same way it reads a credential from any other image. It does not have a special case for the Google name. If the claim generator or a declared action is in the manifest, that is what you see.

Save the original export. A screenshot, a chat-app recompress, or a social download often strips the manifest. After that, the credential reader has nothing to open. Ask for the original file when the decision matters.

SynthID is a different check

SynthID is Google’s invisible watermark for some of its own AI images and video. It is not a C2PA manifest. DeepFakeCheck does not call a SynthID detector and cannot tell you that a watermark is present or absent. Use Google’s own Gemini tools for that watermark. Google’s SynthID overview explains what the watermark is and where Google checks it.

This site also does not extract Digimarc or Adobe TrustMark watermarks, and it does not look up a credential from a watermark after the metadata is gone. If a signer wrote a watermark-related action into a manifest that is still embedded, that action can appear with the rest of the credential. A missing watermark is not used as evidence.

Unsigned text in EXIF or XMP is only a hint. If a metadata field literally contains the word Imagen or SynthID, that string can be shown as an unverified hint. It is not a watermark decode, and it does not change the file into proof.

How to check an image

  1. Keep the original download. Do not start from a screenshot if you can avoid it.
  2. Open the image detector and upload the JPG, PNG, or WebP file. There is no signup. A limited free allowance applies.
  3. Read the credential separately from the visual and pixel notes. One weak or missing layer is not proof the photo is real.
  4. For the SynthID watermark, use Google’s own check. Come back here for the embedded credential and the other signals this detector actually has.

Analysis content is retained for up to 7 days to improve detection quality, then deleted. A high-risk result does not name Nano Banana. A low-risk result does not mean the image is authentic. Ordinary crops and color edits are outside what this detector is built to identify.

Nano Banana detector FAQ

Is this a Nano Banana detector?

It is a guide to the image detector, which is the check to use for a Nano Banana or Gemini image. Upload the original JPG, PNG, or WebP file. The result is screening evidence. It does not certify that the picture, or the event it shows, is real.

Can it tell if an image was made with Gemini?

Only when the file still contains evidence the checker can read. An embedded C2PA Content Credential can name the claim generator and the actions the signer recorded. The visual review looks for general generation or editing signals and does not name Nano Banana, Gemini, or any other product.

Does DeepFakeCheck read Google SynthID?

No. SynthID is Google’s invisible watermark. Google’s own Gemini tools are what check it. This site does not call a SynthID detector. A metadata field that merely contains the word SynthID or Imagen is an unsigned hint, not a watermark decode.

Will it read a C2PA credential on a Gemini or Nano Banana image?

Yes, if that credential is still embedded in the image file. The browser reads the manifest locally, checks its validation state, and can show the claim generator, signer, and declared actions, including a generative-AI source type when the signer recorded one. Google has said some Gemini and Nano Banana images carry this kind of credential. A screenshot or a social-app copy often removes it.

Does it detect Digimarc or Adobe TrustMark watermarks?

No. The checker does not extract a watermark from the pixels and does not look up a credential that was stripped from the file. If a manifest is present, it can report an action the signer already wrote into that manifest. A missing watermark is not treated as evidence.

What file should I upload, and what are the limits?

Use the original export when you have it: JPG, PNG, or WebP, up to 50 MB. No signup is required. A limited free allowance applies. Analysis content is retained for up to 7 days, then deleted. Open the image detector to run the check.

Upload the image