Is this image real or AI? The 60-second check

Why your eyes stopped being enough

Google released Nano Banana Pro (its Gemini 3 Pro Image model) on 20 November 2025. It generates images at up to 4K, renders clean readable text inside the image, and fixed most of the old giveaways like melted hands and gibberish signage. That is genuinely new: the "just look for six fingers" advice is basically dead.

So the move is not to squint harder. It is to have a repeatable check you run every time, the same way you would fact-check a wild headline before resharing it.

Source: Google's Nano Banana Pro announcement

The 60-second check (run all three)

1. 🔁 Reverse image search it first. Before anything else, find out where the image actually came from. Drop it into Google Lens (camera icon in the Google search bar, upload the file) or TinEye, which is built specifically for finding the earliest copy and is great for tracing an original. If the "breaking news" photo traces back to a 2019 stock library or a meme account, you are done in 10 seconds.
2. 🧠 Ask Claude to read the image for inconsistencies. Open Claude, paste or upload the image (Claude accepts JPEG, PNG, GIF, and WebP), and ask it to point out anything physically off. Claude is strong at visual reasoning, warped text, impossible reflections, lighting that does not match, jewelry or fingers that merge. Important honesty note: Claude is a reasoning helper, not a certified detector. It gives you a smart second opinion, not a verdict.
3. 🕵️ Trace the source and cross-check. Ask Claude to help you work out who posted it first and whether any credible outlet is reporting the same event. A real news moment shows up in more than one trustworthy place. One anonymous account with no corroboration is a red flag on its own, regardless of how clean the image looks.

The exact prompt to paste into Claude

I'm trying to work out if this image is real or AI-generated. Analyse it and give me:
1. Physical inconsistencies: warped or nonsensical text, impossible reflections or shadows, lighting or perspective that does not add up, hands, teeth, ears, or jewelry that merge or distort, repeating background patterns.
2. Context clues: does anything in the scene (signs, clothing, weather, tech) look anachronistic or staged?
3. Your honest confidence level (low / medium / high) that this is AI-generated, and exactly what made you lean that way.
4. Three specific things I should go verify myself (who posted it first, whether a credible outlet reports it, what a reverse image search would tell me).
Be clear that you are giving a reasoned opinion, not a definitive detection.

What the detectors actually tell you

Detectors are useful, but only if you know what a result means.

1. 🏷️ SynthID (Google). Every image made by Google's AI tools, including Nano Banana Pro, carries an invisible SynthID watermark. Google opened a SynthID Detector portal at I/O on 19 May 2026 (early access, join the waitlist) that scans an upload and highlights where a watermark is found. A positive result is strong: it was made with Google AI. A negative result proves nothing, because it may have been made with a non-Google tool, or the watermark was stripped by a screenshot or heavy edit.
2. 📜 Content Credentials (C2PA). This is the open provenance standard (now ISO/IEC 22144) that some cameras, Adobe, and OpenAI attach to files, a signed record of what made or edited an image. You can inspect it at contentcredentials.org/verify. Same rule applies: credentials present is informative, credentials absent is inconclusive, because screenshots and re-uploads routinely strip them.

Sources: SynthID Detector portal · What Content Credentials are

3 bonus prompts to run next

🎯 Spot-check a screenshot of a "quote":

Here is a screenshot of a post or headline. Before I believe it, tell me what would make this easy to fake, what specific details I should independently verify, and draft the exact search I should run to confirm or debunk it.

🧾 Build me a personal verify checklist:

Turn the check we just did into a short reusable checklist I can run on any image or viral claim in under 2 minutes, ordered from fastest to slowest step, so I never have to think about it again.

📰 Group-chat translator:

Someone sent me this image claiming [what they claim]. Write me a calm, non-preachy reply I can send back that shares what I found and how I checked it, without making them feel dumb for sharing it.

How to get the most out of it

1. 🌱 If you are new to this: do not skip step one. Reverse image search alone catches a huge share of fakes and recycled photos in seconds, before you even open Claude. It is the cheapest, fastest win.
2. 🚀 If you want to go deeper: save the main prompt as a reusable prompt or a Claude Project so every image gets the same rigorous check. Pair it with the SynthID Detector for anything that might be Google-made, and check Content Credentials on files from newsrooms or Adobe workflows.

An honest heads-up