"Nano Banana" is the nickname that stuck to Google's Gemini image models — first the Flash image generator, then the sharper Nano Banana Pro. Where a tool like Grok ships almost no provenance, Google went the opposite direction and tags its output three different ways. Pictures made on a free account usually get a small visible glyph in the corner. Every image, free or paid, carries SynthID — an invisible watermark written into the pixels themselves. And on top of that sits a C2PA "Content Credentials" record, cryptographically signed by Google, that says in plain terms that the image is AI-generated.
That's good news: for once, "is this a Gemini image?" is a question you can often answer directly instead of guessing from the pixels. The catch is that each of the three marks fails in a different way — the visible one crops off, the metadata strips on upload, and the invisible one only reads if Google's own tools get to look. So you check them in order, and you read the result for what it is. Here's how I'd work through it.
The 6-step check
Look for the visible mark first
Images that come out of the Gemini app on a free account — Nano Banana and the newer Nano Banana Pro — usually carry a small visible watermark in a bottom corner, a little sparkle/diamond glyph that tags the picture as Google AI. That's the fastest tell when it survives. Open the image at full size and scan the corners. Two catches, though: a corner glyph sits in one fixed spot, so it's trivial to crop, and paid or API-generated Nano Banana images often don't carry the visible mark at all. Present is a useful hint; absent tells you almost nothing on its own.
Ask Gemini whether it made the image
This is the step that's actually specific to Google's models. Every image Nano Banana generates carries SynthID — an invisible watermark baked into the pixels, not the metadata, so it survives screenshotting, resizing, mild cropping, and re-compression that would wipe ordinary EXIF. Upload the picture to the Gemini app and ask "Was this created with Google AI?" Gemini checks for the SynthID mark and tells you what it found. Google says this in-app check has been used tens of millions of times since it launched in late 2025. It won't flag a Midjourney or Stable Diffusion image — SynthID only rides on Google's own models and a few partners (OpenAI began embedding it in 2026) — but for the question "is this Nano Banana / Gemini?" it's the most direct answer you can get.
Run it through the SynthID Detector or Google Lens
You don't have to stay inside the chat app. Google opened a standalone SynthID Detector portal that scans an uploaded image and highlights which regions carry the watermark, and in 2026 it rolled the same check into Google Lens, AI Mode, and Circle to Search — point any of those at an image and it reports whether it finds a SynthID watermark or C2PA credentials. Gemini in Chrome also lets you right-click an image on a page and check it in place. Use whichever is in front of you; they're reading the same underlying signal.
Inspect the C2PA Content Credentials
Alongside SynthID, Nano Banana images are stamped with C2PA "Content Credentials" — a provenance manifest cryptographically signed by Google LLC that records the tool used and the fact that the image is AI-generated. Any open C2PA viewer reads it: drop the file into the Content Authenticity Initiative's checker at contentcredentials.org/verify, or use Google's "About this image." One honest limitation: metadata is fragile. Screenshotting an image, or an upload pipeline that re-encodes it, often strips the C2PA manifest entirely — which is exactly why SynthID (step 2) matters, because the pixel watermark can outlive the metadata.
Read the image for the usual AI tells
If the markers are gone — cropped out, stripped, or degraded past detection — fall back on what the picture shows. Nano Banana is very good at photoreal scenes and notably better at legible text than older models, so the classic "the signs spell gibberish" tell is weaker here than it used to be. Keep looking anyway at the fiddly bits: hands with an extra or fused finger, jewelry and teeth that merge, earrings or eyes that don't match, backgrounds where railings bend or crowds blur into paste, and lighting that lands on faces from directions the scene can't explain. A too-perfect, faintly plastic micro-texture across skin and fabric is another. None of these settles it alone — but a cluster of them on an image making a big claim is your cue to slow down.
Verify the claim, not just the pixels
Pinning a file to Nano Banana answers "how was this made" — but the harm usually rides on the story attached to it: a supposed event, a person shown doing something, a "leaked" scene. That part you can check directly. Reverse image search a copy to find the earliest version or the source photo it was built from, look for whether any credible outlet has reported the event, and see whether fact-checkers have already weighed in. That last one is quick: FAXTR searches 100+ fact-checking organizations in a single query, so you can find out whether the image in your feed already has a published verdict before you pass it on.
Signals worth training your eye on
You won't run a full SynthID pass on every image that crosses your feed. These are the tells worth catching at a glance — any one of them is reason enough to look closer before you believe or forward it.
- ✕A small sparkle/diamond glyph tucked into a bottom corner — the visible Gemini mark on free-tier Nano Banana images.
- ✕A blurred, cropped, or cloned corner where that glyph could have been trimmed off.
- ✕Gemini, the SynthID Detector, or Google Lens reporting a SynthID watermark when you ask whether it's AI.
- ✕C2PA Content Credentials signed by "Google LLC" visible in contentcredentials.org/verify or "About this image."
- ✕Photoreal skin and fabric with an oddly uniform, plastic micro-texture and no sensor grain.
- ✕A dramatic image that no news outlet has published and that exists only as this one file.
Why a "clean" image still might be Nano Banana
It feels logical that a real Gemini image would carry its mark, so a picture with no visible glyph and no credentials must be a genuine photo. Don't lean on that. The corner glyph is a crop away from gone, and the C2PA manifest rarely survives a screenshot or a trip through a social platform's re-encoder — so the most convincing fakes, the ones cleaned up and reshared until they look native to your feed, are exactly the ones with the obvious markers stripped. The saving grace is SynthID: because it lives in the pixels, it can outlast a screenshot and a crop, which is why step 2 is worth doing even when steps 1 and 4 come up empty. But flip the logic too — a SynthID "no detection" result does not mean "real photo." It means "not made by Google or a SynthID partner," which still leaves Midjourney, Stable Diffusion, Firefly, and a plain camera all on the table.
What this guide is not
This isn't a claim that Nano Banana images are inherently deceptive — most are people making art, mockups, and jokes, posted plainly as such. And none of these checks is a lab result: SynthID is strong but not infinite (enough edits will eventually scrub it), and a detector reading on the raw pixels is a probability, not a verdict. The point is a repeatable habit for the moment a "real" photo makes a big claim — read the visible mark, ask Gemini about the invisible one, inspect the credentials, and check the story either way before it travels any further on your say-so.
Check the claim in seconds
A viral image making a big claim? FAXTR searches 100+ fact-checking organizations in one query — free, no login required.