How to Read SynthID Watermarks and C2PA Credentials on AI-Generated Images

You’ve been generating images with Nano Banana for a while now. Great. But here’s something most users skip entirely: every single image that comes out of Google’s Gemini-powered image generator is tagged twice — once with a SynthID watermark baked into the pixels themselves, and once with a C2PA credential packet attached to the file metadata. These aren’t just legal cover for Google. They’re genuinely useful tools if you know how to read them — for verifying your own work, crediting AI-generated content correctly, and not getting caught out claiming a photo is something it isn’t.

This tutorial covers both systems from scratch: what they are, how they work in Nano Banana specifically, how to verify a Nano Banana image in the wild, and how to think about disclosure when you’re publishing. No fluff, just the mechanics.

What SynthID Actually Does (and Doesn’t Do)

SynthID is DeepMind’s steganographic watermarking system. The key word is steganographic — the watermark isn’t a visible logo stamped in the corner. It’s a signal embedded directly into the pixel values of the image at generation time, statistically imperceptible to the human eye but detectable by the SynthID verification model. Google has used SynthID across Imagen outputs since 2023, and it carries over into Nano Banana images generated via the Gemini app, AI Studio, and the Gemini API.

What SynthID does: it lets Google’s own verification tool confirm whether a given image was generated by a Google AI system. What it doesn’t do: it doesn’t survive aggressive JPEG recompression, heavy cropping, or significant color grading. It also doesn’t tell you who generated the image — just that a Google model generated it. Think of it as a provenance signal, not a tracking device. If you export from Nano Banana and immediately run it through three Instagram filters and a lossy resize, the SynthID signal may degrade. If you export at full resolution and keep the file intact, it’ll hold.

Pro tip ✅

Always export your Nano Banana images at the highest available resolution and keep an uncompressed master copy before resizing for social. The SynthID watermark survives normal web export (standard JPEG compression), but aggressive editing pipelines can weaken or destroy it. Your master file is your source of truth.

What C2PA Credentials Are and Why They Matter More for Publishing

C2PA (Coalition for Content Provenance and Authenticity) is an open technical standard, backed by Adobe, Microsoft, Google, and a consortium of news organizations, for attaching cryptographically signed provenance metadata to media files. Where SynthID is invisible and pixel-level, C2PA credentials live in the file’s metadata — they’re a signed manifest that records what created the content, when, and what assertions the creator is making about it.

Nano Banana images carry a C2PA manifest asserting that the content was AI-generated by a Google system. This is readable by any C2PA-compatible tool, and it’s what news organizations and stock photo platforms increasingly check before accepting images. The Adobe Content Authenticity Initiative’s free verification tool (verify.contentauthenticity.org) reads C2PA credentials directly — you upload an image and it shows you the full provenance chain. Google is a C2PA member, so Nano Banana’s credentials use the standard format that integrates with Adobe’s ecosystem.

Note 💡

C2PA credentials are stored in file metadata, which means they will be stripped by most social media platforms when they re-process your upload. Twitter/X, Instagram, and TikTok all strip metadata on ingest. If you need the C2PA trail to survive, share the file directly (via Google Drive, Dropbox, or as an email attachment) rather than uploading to a platform that processes images.

How to Verify a Nano Banana Image Step by Step

The verification workflow is straightforward. You’ll use two tools: Google’s SynthID verification (available via the About this image feature in Google Search and Google Lens) and the C2PA verification at verify.contentauthenticity.org.

Step 1 — Verify SynthID via Google Lens. Open Google Lens in your browser or the Google app. Upload or paste the Nano Banana image. Tap the three-dot menu and select “About this image.” If SynthID detects a Google AI watermark, you’ll see an AI-generated content disclosure in the results panel. This works on images that have kept their pixel data intact.

Step 2 — Verify C2PA credentials. Go to verify.contentauthenticity.org — this is Adobe’s free public tool, and it reads C2PA from any compliant image. Drop your Nano Banana image into the upload zone. The tool will display the content credentials manifest: who signed it (Google), the AI model assertion, and any editing history that’s been appended since generation. If the file was exported cleanly from Nano Banana and not stripped of metadata, you’ll see a green “Content Credentials” badge with the full chain.

Step 3 — Check within AI Studio or the Gemini API. If you generated the image through AI Studio or the Gemini API programmatically, the API response includes generation metadata you can log server-side. This is useful for developers building pipelines: store the generation timestamp, model ID, and prompt hash alongside the image file so you have provenance documentation independent of what survives in the file itself.

Pro tip ✅

If you’re building a Nano Banana workflow via the Gemini API, log the full API response object for every image generation call. The metadata in the response is more reliable than what survives in the exported file, especially if your pipeline includes any image processing steps. A simple database entry with image hash + generation timestamp + model ID is solid provenance documentation.

Prompts to Test SynthID and C2PA Behavior

These prompts are designed to produce clean, verifiable Nano Banana outputs — images with intact SynthID and C2PA credentials that you can run through the verification workflow above. Use them as test cases when you’re setting up a new pipeline or checking that your export settings preserve metadata.

A product photograph of a glass perfume bottle on a white marble surface, soft studio lighting, no text, no watermarks, 4K resolution, photorealistic

This produces a clean, high-resolution product image with no overlaid text that could confuse the SynthID signal. The “no watermarks” instruction is a prompt-level note to Nano Banana’s text renderer — it doesn’t affect SynthID, which is at the pixel level. Good baseline test image.

An aerial photograph of a dense tropical rainforest canopy at golden hour, cinematic color grading, ultra-detailed, 4K

Nature scene with no faces and no text — clean canvas for testing whether your export pipeline preserves the SynthID signal after a standard JPEG save. Compare the verification result before and after your usual resize operation.

Editorial portrait of a woman in a red blazer sitting at a minimalist desk, natural window light, shallow depth of field, photorealistic

Portrait prompt useful for testing C2PA credential display in newsroom CMSs. Many editorial content systems now have C2PA readers built in at upload. Drop this export into your CMS’s media uploader and check whether it flags the AI-generated assertion.

A vintage-style travel poster for an imaginary mountain city called "Alto Verde", hand-lettered title text, illustrated style, warm color palette

This one includes text rendering, which is where Nano Banana genuinely earns its keep. The generated image will have C2PA credentials asserting AI generation even though the style looks hand-crafted. Good demonstration case for the gap between visual style and provenance metadata.

Close-up photograph of a fresh espresso shot in a white ceramic cup, coffee crema detail, dark wooden surface, natural light, 4K, photorealistic

Food/beverage photography prompt. Useful for testing whether stock platforms (Shutterstock, Adobe Stock) correctly reject or flag the C2PA-credentialed file. Spoiler: both platforms now check for C2PA AI assertions and will either reject or re-categorize the submission automatically.

Architectural rendering of a modernist house with floor-to-ceiling windows, surrounded by pine trees, overcast sky, photorealistic, 4K

Architecture render — a common commercial use case. If you’re a designer using Nano Banana to mock up concepts, this is the category where C2PA documentation matters most for client contracts. The credential chain proves the image is AI-generated, not a photograph of an existing building, which matters for rights clearance.

A flat lay of stationery items — notebooks, pens, a small plant — on a light grey surface, overhead shot, minimalist, clean product photography style

E-commerce flat lay. Tests the full workflow: generate, export full-resolution, verify SynthID via Lens, check C2PA via verify.contentauthenticity.org, then use the image in a product listing with proper AI-generated content disclosure.

A photorealistic still life painting style image of seasonal fruit arranged on a dark wooden table, dramatic chiaroscuro lighting, museum quality

Art-adjacent prompt that produces an image visually indistinguishable from a photograph of a painting — a genuinely tricky provenance case. The C2PA credential will correctly assert AI generation regardless of the visual style. This is exactly the scenario C2PA was designed for.

Warning ⚠️

Don’t assume SynthID verification failure means an image isn’t AI-generated. A negative SynthID result means either the image wasn’t made by a Google system, or the watermark was degraded. It’s not a clean bill of authenticity. Use C2PA as your primary documentation standard for publishing workflows — it’s more robust to normal editing operations than the pixel-level SynthID signal.

Disclosure Best Practices When Publishing Nano Banana Images

The EU AI Act requires disclosure for AI-generated images used in commercial contexts. Several major style publications now require C2PA credentials on submitted photography. The practical answer is simple: keep the file metadata intact when submitting to publications or agencies, and add a text disclosure (“Generated with AI / Google Gemini”) in your caption or alt text as a belt-and-suspenders approach. The C2PA credential handles the machine-readable layer; the text disclosure handles the human-readable one.

For social media, where C2PA gets stripped on upload, platform-level AI disclosure labels are increasingly mandatory — Meta, YouTube, and TikTok all have AI content labeling policies that require self-disclosure. Tagging your Nano Banana images as AI-generated in the platform’s own disclosure UI is both a compliance step and, frankly, the honest thing to do. The SynthID signal surviving in Google Search’s “About this image” feature means Google’s systems may surface the AI-generated label in search results anyway, even if the platform strips the C2PA metadata.

Pro tip ✅

If you’re running a commercial workflow — social media management, e-commerce, editorial — set up a simple naming convention for your Nano Banana exports: NB2_[date]_[project]_MASTER.jpg for the full-res unprocessed export, and NB2_[date]_[project]_WEB.jpg for the compressed version. Keep the master on file. If anyone ever questions the provenance of a published image, you have the metadata-intact master to run through C2PA verification.

Using Vertex AI and AI Studio for Provenance-Tracked Pipelines

If you’re accessing Nano Banana through the Gemini API via AI Studio or Vertex AI rather than the consumer Gemini app, you have more control over provenance documentation. The API response for image generation includes model metadata and generation parameters that you can store alongside the image. Vertex AI adds enterprise-level audit logging, which records generation events with timestamps and project IDs — useful for organizations that need to demonstrate compliance with AI content policies to clients or regulators.

In AI Studio, generated images can be inspected directly in the interface before export — Google’s own UI shows the C2PA badge when you hover over a generated image, giving you a quick sanity check that credentials are attached before you download. This is the fastest way to confirm the image came out of the generation process with intact provenance before it enters any downstream workflow.

Note 💡

Vertex AI’s audit logs and AI Studio’s provenance UI are only as useful as your downstream documentation practice. Build provenance logging into your pipeline at generation time, not as an afterthought. Retroactively reconstructing who generated what image, when, for which project, from which prompt — is a painful exercise that C2PA and good naming conventions can entirely prevent.

Why This Actually Matters Now

SynthID and C2PA credentials were easy to ignore when AI image generation was a novelty. They’re harder to ignore now that stock agencies are checking for them, publications are requiring them, and regulators are writing laws around AI content disclosure. Nano Banana attaches both automatically — you don’t have to do anything to get the watermark or the credentials. The work is in preserving them through your pipeline and using the verification tools when you need to prove provenance in either direction (confirming your own work is AI-generated, or checking whether someone else’s image is). The tools are free, the standard is open, and the metadata is already there waiting to be read. The only question is whether you’re actually using it.

promptyze

See Full Bio