Gemini watermark: what Nano Banana adds and what comes off
Gemini images carry a visible sparkle, invisible SynthID and C2PA metadata. Which mark does what, which you can remove, and an honest look at remover tools.
4 min read
On this page7 sections
Short answer. A Gemini or Nano Banana image carries up to three marks. The sparkle logo in the corner is drawn into the pixels. SynthID is an invisible watermark in the pixels that is built to survive edits. The metadata (C2PA Content Credentials and IPTC AI fields) is what Instagram and TikTok read to label the post. Metadata is the layer you can remove reliably; HideSlop's image remover does that and re-saves the image as an iPhone photo. The sparkle needs an edit, and no tool can honestly promise to remove SynthID.
The three layers
| Mark | Where it is | Can you see it? | Who reads it | Removable? |
|---|---|---|---|---|
| Sparkle logo | Bottom corner of the picture | Yes | People | By cropping or editing the pixels |
| SynthID | Spread across the pixels | No | Google's SynthID Detector | Not reliably; see below |
| C2PA and IPTC metadata | The file container, next to the image | No | Instagram, Facebook, Threads, TikTok, LinkedIn, Google | Yes |
Google added the layers at different times. SynthID has been in Imagen output since 2023. The visible sparkle appeared when image generation came to the Gemini app. Content Credentials followed when Google joined the C2PA effort and started signing Gemini output. Nano Banana (the image model inside Gemini) and Nano Banana Pro write all three in the consumer app.
Why Nano Banana has a watermark
Google's position is that AI images should be identifiable, and it uses three mechanisms because each fails in a different way. The sparkle is for people scrolling a feed. Metadata is for platforms and tools. SynthID is the backstop for when the other two are stripped, because a screenshot removes both the metadata and, if you crop, the logo.
Google has said that some paid tiers and developer access produce images without the visible sparkle while keeping SynthID and the metadata. Which plans qualify changes, so check the current terms for your subscription or the API rather than relying on a guide.
What the metadata contains
A Gemini export typically carries a C2PA manifest signed by Google, with an action of "created" and a digital source type of trained algorithmic media, plus XMP fields that name the generator. That is exactly what Meta and TikTok look for. You can see it in seconds with the free AI label checker, which runs on your device.
What you can remove
The metadata
The image remover rebuilds the file from the pixels. The manifest, the IPTC field and the XMP names are gone, and camera metadata from a real iPhone model is written in their place. The image is saved as a JPEG at phone quality with an invisible variation (a one to five percent zoom, a slight color shift, grain) so the file is also new to duplicate matching. Up to 20 images at once, first one free.
After this, Instagram and TikTok have nothing in the file to label. The sparkle, if your tier adds one, is still in the picture.
The sparkle
The logo is part of the image. The options are the ones you would use for any logo: crop it out when the composition allows, or paint over it with a clone or remove tool in an editor. Both change pixels, and both leave SynthID in place, because SynthID is spread across the whole image and does not live in the corner.
SynthID
Google designed SynthID to survive cropping, compression, resizing, filters and screenshots. Sites that call themselves SynthID removers cannot show that they remove it, and the only test is Google's own detector. HideSlop changes every pixel slightly, which is enough to defeat ordinary fingerprint matching, but it makes no claim about SynthID. The full picture is in can you remove SynthID?
What this means for posting
For the automatic AI labels on Instagram, Facebook, Threads and TikTok, the metadata is the trigger, and it is removable. For the sparkle, generate in a tier that does not draw one or compose with the corner in mind. For SynthID, assume it is there. Platforms do not currently label posts based on SynthID, but Google can detect it, and that may change.
If your content is realistic, platform rules ask you to disclose it regardless of what the file says.
Checking what is left
- Metadata: the AI label checker shows Content Credentials, the AI source type and generator names, and confirms a clean file.
- SynthID: upload the image to Google's SynthID Detector, or ask the Gemini app whether an image was made with Google AI.
FAQ
Is there a free Gemini watermark remover?
For the metadata, yes: the first removal on HideSlop is free. For the visible sparkle, any editor with a crop or remove tool. For SynthID, no tool has shown it can remove it, free or paid.
Does removing the sparkle remove SynthID?
No. The sparkle is in one corner; SynthID is across the whole image. They are independent.
Does Instagram label Gemini images?
Yes, when the file carries Google's Content Credentials or the IPTC AI field. With the metadata removed, there is no automatic label.
Are Nano Banana Pro images watermark-free?
Google has said some paid tiers skip the visible mark while keeping SynthID and the metadata. Check your plan's current terms; the policy has changed before.
Does a screenshot remove the Gemini watermark?
A screenshot removes the metadata and, if you crop, the logo. SynthID is designed to survive screenshots.