Why Nano Banana Watermarks Stay: SynthID Explained (April 2026)

A common question in the Gemini Apps Community and across Reddit goes something like: "I'm paying for Gemini Pro, why do my Nano Banana images still have a watermark?" The answer is more interesting than "Google forgot to remove it." There are actually two different things called a watermark on Nano Banana output, and only one of them is the visible Gemini sparkle in the corner.

This guide separates the visible logo from the invisible SynthID signature, explains what Google DeepMind has documented about each, and lays out why both persist across Nano Banana, Nano Banana Pro, and Nano Banana 2 (Gemini 3.1 Flash Image). It does not teach removal. It explains the policy, technical, and regulatory reasons the marks are there.

TL;DR

• Two watermarks, not one. A visible Gemini sparkle logo on consumer-tier downloads, plus an invisible SynthID signal embedded into the pixels. Different systems, different purposes.
• SynthID is invisible by design. Google DeepMind states the watermark "doesn't change the image or video quality" and is added "the moment content is created" (source).
• It survives common edits like "cropping, adding filters, changing frame rates, or lossy compression" (source).
• It is now a regulatory requirement. EU AI Act Article 50 and the Draft Code of Practice (Dec 2025) require providers to mark AI-generated content in machine-readable form and to "preserve marks" against removal (source).
• Pro tier doesn't strip it. Per Google DeepMind, "All of our images are imperceptibly watermarked with SynthID technology" (source).

The two-watermark stack on a Nano Banana image

A Nano Banana download can carry up to three layers of provenance signal stacked on top of one another:

These are routinely conflated in user discussions. They are not the same. Cropping the corner logo does nothing to the SynthID signal embedded across the rest of the pixels, and re-saving the file to strip C2PA metadata does nothing to either of the other two.

What SynthID actually is

Google DeepMind describes SynthID as "our new watermarking tool, designed specifically for AI-generated content. It empowers users to identify AI-generated (or altered) content, helping to foster transparency and trust in generative AI" (source).

For images, SynthID is not a logo and not metadata. DeepMind's research blog explains the technical approach: SynthID uses "two deep learning models — for watermarking and identifying — that have been trained together on a diverse set of images," optimized for "correctly identifying watermarked content and improving imperceptibility by visually aligning the watermark to the original content" (source).

In plain terms: a neural network nudges pixel values across the entire image in a pattern another network is trained to recognize, calibrated to stay under the human visual threshold. The result is a hologram-like signal distributed across the image rather than concentrated in any single corner, which is why the watermark survives cropping. Per DeepMind, it persists "even after modifications like adding filters, changing colours, and saving with various lossy compression schemes, most commonly used for JPEGs" (source).

Why it persists across Nano Banana, Pro, and Flash

Google's product page for Gemini 3 Pro Image (Nano Banana Pro) states: "All of our images are imperceptibly watermarked with SynthID technology, allowing you to detect whether an image has been created or edited using AI."

The phrase "all of our images" is the operative one. SynthID embedding is part of the model serving pipeline, not a per-tier setting flipped on for free users. The same applies to Nano Banana 2 (Gemini 3.1 Flash Image) on the Flash variant page, which Google says includes "all the latest privacy and safety features, including SynthID."

This is why upgrading to Google AI Pro does not remove the SynthID layer. The Pro tier changes quotas, resolution ceilings, and (per third-party reporting) whether the visible Gemini logo is composited on top, but not whether the SynthID neural watermarker runs. That step is upstream of any tier configuration. The visible Gemini sparkle logo is a separate, tier-dependent overlay; the Gemini Apps Community thread on it is where most user confusion gets debated.

Claimed vs confirmed

Separating what Google has documented from what third parties have verified:

Claimed by Google DeepMind:

• Imperceptible to the human eye (source).
• Survives cropping, filtering, color changes, and JPEG compression (source).
• Used to watermark "over 10 billion pieces of content" across Google services (source).
• Still detectable even when file metadata is stripped.

Independently studied:

• The paper SynthID-Image: Image watermarking at internet scale on arXiv presents evaluations at the deployment scale Google describes.
• Detector outputs use thresholds, not yes/no certainty: "Detected," "Not Detected," and "Possibly Detected" are the published outcomes.
• DeepMind acknowledges the limit: "SynthID isn't foolproof against extreme image manipulations, but it does provide a promising technical approach" (source).

Imperceptibility and resilience to standard edits are well documented. Resilience against deliberate, targeted attacks is acknowledged as not foolproof.

Why the watermark exists: ethics, policy, and now law

Three forces keep SynthID on every Nano Banana image, and they are reinforcing each other.

1. Provenance for the open internet

Generative image models produce photorealistic images of people, events, and products that never existed. Without an embedded provenance signal, downstream consumers (search engines, news outlets, courts, voters) have no automated way to ask "was this made by an AI?" Pixel-level marks like SynthID and signed metadata like C2PA Content Credentials are how that question gets answered at scale.

2. The C2PA / Content Credentials coalition

Google joined the Coalition for Content Provenance and Authenticity (C2PA) as a steering committee member alongside Adobe, BBC, Intel, Microsoft, Sony, Truepic, and Publicis Groupe. In its own blog post, Google described SynthID as a complement to C2PA: pixel-embedded watermarking that survives modifications, plus signed metadata that creates verifiable attribution. Per the same post, Nano Banana Pro images now ship with C2PA metadata in addition to SynthID.

The two signals are designed to work together. SynthID survives the operations that strip metadata (re-saving, screenshotting, format conversion). C2PA carries richer human-readable provenance (which model, when, what edits) that an algorithmic watermark cannot encode. Removing one does not remove the other; that is the point.

3. The EU AI Act and the Draft Code of Practice

The strongest force is now legal. Article 50 of the EU AI Act requires providers of generative systems to ensure outputs "are marked in a machine-readable format that identifies the content as artificially generated or manipulated."

The first draft of the Code of Practice was published 17 December 2025 and is expected to be finalized in May–June 2026, ahead of the August 2026 enforcement deadline (source). Penalties reach €15 million or 3% of global annual turnover, whichever is higher.

Two clauses in the Draft Code matter for why Nano Banana watermarks "stay":

• Imperceptible watermarking is required. The Code envisages "imperceptible watermarking that is 'interwoven' with the content," durable against compression and cropping. That language maps onto how SynthID is built.
• Providers must preserve the marks. Providers "shall ensure and implement appropriate measures to preserve marks … by technically ensuring that existing detectable marks are retained and not altered or removed," and must "contractually ensure prohibition for removal or tampering of the marks by deployers or any other third parties."

The law doesn't just say "watermark your images." It says "design watermarks that survive modification, deploy them, and contractually forbid removal." That is the architecture Google has built around Nano Banana.

Visible logo vs invisible signature, and what the Terms say

The Reddit and Gemini Help question ("I'm paying, why is the watermark still there?") almost always means the visible Gemini sparkle. Per third-party documentation of the consumer experience, the visible logo behavior varies by tier (free, Google AI Pro, Google AI Ultra) and by surface (Gemini app vs API). That is a branding choice. The SynthID layer is different: it runs on every output (Pro, Flash, free, paid, app, API) because it is the provenance commitment baked into how DeepMind ships these models. There is no "Pro tier strips SynthID" toggle, and the EU AI Act Code of Practice now prohibits Google from offering one to deployers.

This article does not endorse, document, or teach watermark removal. On the legal side, the Gemini API Additional Terms of Service governs commercial use of generated content and requires users to comply with applicable law. The EU Code of Practice draft requires providers to contractually prohibit "removal or tampering of the marks by deployers or any other third parties." For commercial use, the conservative path is to read the current Gemini API Terms and current Nano Banana Pro product page, and treat both the visible mark and the SynthID layer as part of the deliverable unless your license explicitly says otherwise.

Detection: the SynthID Detector portal

Google has rolled out a SynthID Detector portal that scans uploaded media for the SynthID signature. Per Google, the detector "scan[s] the media for a SynthID watermark" and "highlight[s] specific portions of the content most likely to be watermarked." Access is limited; Google is "starting to roll this out to early testers today, before making it available more broadly," with journalists, media professionals, and researchers able to join a waitlist.

The asymmetry (embed on every output, detect through a controlled portal) is how Google manages the adversarial concern that publishing the detector publicly would make removal easier.

What this means for image creators

For anyone building on Nano Banana or Nano Banana Pro:

1. Assume SynthID is on every output. For most commercial use this is irrelevant. For sensitive contexts (journalism, evidence), be honest about the source.
2. Don't conflate the logo with the signature. Visible overlay vs upstream model behavior are different systems.
3. Plan for C2PA going mainstream. Adobe, Microsoft, OpenAI, Meta, Google, and Amazon are in the coalition; social platforms are starting to display Content Credentials.
4. Track the EU AI Act timeline. August 2026 enforcement of Article 50 locks this architecture in across providers serving the EU.

Frequently asked questions

Why does my Nano Banana image still have a watermark on Gemini Pro?

Two answers, depending on which watermark you mean. The visible Gemini sparkle logo is a tier-dependent overlay in the Gemini app's download flow. The invisible SynthID signature is embedded into the pixels by DeepMind's image models and runs on every generation regardless of tier (source).

Is SynthID actually invisible?

Per DeepMind, yes: "The watermark doesn't change the image or video quality" (source). It uses a neural network to nudge pixel values below the human perception threshold (source).

Does SynthID survive cropping or screenshotting?

Per DeepMind, the watermark is "designed to stand up to modifications like cropping, adding filters, changing frame rates, or lossy compression" (source). DeepMind also acknowledges it "isn't foolproof against extreme image manipulations" (source).

Can I detect SynthID on someone else's image?

A SynthID Detector portal exists but is waitlisted for journalists, media professionals, and researchers (source). General public access has not been announced.

Does removing the visible Gemini logo also remove SynthID?

No. The visible logo is a corner overlay; SynthID is a distributed pixel-level signal across the whole image. Cropping the corner does nothing to the embedded signal.

Is it illegal to remove SynthID?

The EU AI Act's Draft Code of Practice (December 2025) requires providers to "contractually ensure prohibition for removal or tampering of the marks by deployers or any other third parties" (source). In EU jurisdictions, removal can implicate AI Act compliance once enforcement begins in August 2026. This article does not provide guidance on removal.

Does Nano Banana Pro carry C2PA Content Credentials too?

Yes. Nano Banana Pro images ship with C2PA metadata alongside SynthID watermarks (source). C2PA covers signed metadata; SynthID covers pixel-level robustness. Together they form a two-layer provenance stack.

Try Nano Banana through gptimg

The fastest way to see the model family in action is to run a few prompts through gptimg's Nano Banana studio. Output behaves exactly as Google's pipeline does, including the SynthID layer, because the underlying model is Google's, served via API. For the native 4K tier, see Nano Banana Pro. For a head-to-head with OpenAI's flagship, see Nano Banana Pro vs GPT Image.

Sources

• SynthID, Google DeepMind product page
• Identifying AI-generated images with SynthID, Google DeepMind research blog
• Gemini 3 Pro Image (Nano Banana Pro), Google DeepMind
• Gemini 3.1 Flash Image (Nano Banana 2), Google DeepMind
• SynthID Detector portal, Google blog announcement
• How Google and the C2PA are increasing transparency for gen AI content, Google blog
• Google to join C2PA, Coalition for Content Provenance and Authenticity
• SynthID-Image: Image watermarking at internet scale, arXiv research paper
• EU AI Act: First Draft Code of Practice on Transparency and Watermarking Released, Cooley legal analysis
• Gemini API Additional Terms of Service, Google AI for Developers
• Prevent or get rid of Gemini logo on Nano Bananas Pro downloaded image, Gemini Apps Community thread

Last reviewed against source pages: 2026-04-18. SynthID and EU AI Act implementation details change frequently; confirm in the linked sources before acting on the specifics above.