→Claude Just Joined Your Meetings, But Only on Google Meet →An AI Vendor Breach Just Leaked Uber Driver IDs and Notion Chat Logs at the Same Time →A Safety Reminder Cut Harmful AI Medical Advice by a Third. It Still Got 1 in 10 Wrong. →A Power Bank Just Launched That Costs 75 Percent More to Not Catch Fire →Firefox's New Mini Window Is Genuinely Useful, With One Annoying Catch →How to Actually Check if Your Email or Password Was Leaked in a Data Breach →Google's New AI Model Reads DNA Cell by Cell, and the Numbers Are Real →Perplexity's Comet Browser Is Free Now: What You Gain and What You're Trusting It With →Dropbox Just Spent $350 Million on a Storage Startup to Escape Being Dropbox →AI Is Not Taking Your Job. It Might Just Be Freezing Your Raise →NotebookLM Is Gone, Gemini Notebook Is Here: What Actually Changed →Four AI Models, Five Accents, One Hierarchy: What a New Study Found →Claude Just Joined Your Meetings, But Only on Google Meet →An AI Vendor Breach Just Leaked Uber Driver IDs and Notion Chat Logs at the Same Time →A Safety Reminder Cut Harmful AI Medical Advice by a Third. It Still Got 1 in 10 Wrong. →A Power Bank Just Launched That Costs 75 Percent More to Not Catch Fire →Firefox's New Mini Window Is Genuinely Useful, With One Annoying Catch →How to Actually Check if Your Email or Password Was Leaked in a Data Breach →Google's New AI Model Reads DNA Cell by Cell, and the Numbers Are Real →Perplexity's Comet Browser Is Free Now: What You Gain and What You're Trusting It With →Dropbox Just Spent $350 Million on a Storage Startup to Escape Being Dropbox →AI Is Not Taking Your Job. It Might Just Be Freezing Your Raise →NotebookLM Is Gone, Gemini Notebook Is Here: What Actually Changed →Four AI Models, Five Accents, One Hierarchy: What a New Study Found
Panda Tech Bytes
All Posts
Latest AI Trends Guides Tools Tech Pulse Hardware & Gear Funding Watch Risk & Reality What It Means for You Security Watch On Your Desk
How to Actually Check if a Photo Is AI-Generated (Not by Staring at Hands)

How to Actually Check if a Photo Is AI-Generated (Not by Staring at Hands)

Counting fingers and checking shadows used to work, but modern AI images are clean enough that those tricks mostly fail now. Here is how to use the actual provenance tools, SynthID and Content Credentials, that can tell you for real.

For a while, the advice for spotting a fake photo was to zoom in on the hands, check the shadows, and look for weird text in the background. That advice is getting less useful every month. The newest image models are clean enough that those tells are increasingly rare, which is exactly why Google and others have shifted the whole approach: instead of asking you to eyeball an image, they built tools that check the image's actual digital history. That is the method worth learning, because it is the one that still works when the visual clues do not.

Why the old tricks are losing their edge

Hands with six fingers, mismatched earrings, and text that dissolves into gibberish were reliable signs of AI generation for years. They still show up sometimes, especially from older or lower-end tools. But current-generation models handle hands, reflections, and fine text far better than they did even a year ago, and some are specifically trained to avoid those exact giveaways because so many people learned to look for them.

That does not mean visual inspection is worthless. Reverse image searching a photo, checking whether reflections in mirrors or water match the scene, and noticing when perspective lines do not quite converge can still catch sloppy fakes. The problem is that it only catches the sloppy ones. A careful fake, or a well-made one, will sail right past a visual check.

The real tool: SynthID and the Gemini app

Google's SynthID is an invisible watermark baked into images (and now video and audio) made or edited by Google's AI tools, including Gemini's image generation and Imagen. It survives cropping, resizing, filters, and even screenshots, which is what makes it different from a visible logo you could just crop out.

You can check for it yourself, for free, right now:

According to Google, this verification feature has already been used roughly 50 million times since it rolled out, and more than 20 billion pieces of AI-generated content have been watermarked with SynthID since the technology launched in 2023. Google has also said it plans to bring this same checking ability into Search and Chrome, so you would not need to open a separate app to run the check.

The catch: it only works on content Google can see

SynthID can only flag an image if it was made or edited with a tool that actually embeds the watermark. That means Gemini, Imagen, and other Google AI tools. If someone generates a photorealistic image with Midjourney, an open-source model, or a tool that strips out watermark data, there is nothing for Gemini to detect, and asking it will not give you a reliable answer.

This is the most important limitation to understand before you rely on this method: a "no SynthID detected" result does not mean an image is real. It only means Google did not find its own watermark. Treat it as one useful signal, not a verdict.

The other half of the puzzle: Content Credentials (C2PA)

A separate but related system, called Content Credentials and built on a standard known as C2PA, works the opposite way: instead of flagging AI-made content, it certifies how a photo was actually captured or edited, attaching metadata about the camera, software, and edits involved.

Camera-side adoption is uneven. Google's Pixel 10 attaches Content Credentials to every photo taken with its camera app by default, backed by the phone's hardware. Samsung's Galaxy S25 only attaches credentials to images that were created or edited using generative AI features, not to ordinary photos. No iPhone currently supports Content Credentials at all, and Apple has not committed to adopting the standard. On the professional side, camera makers including Sony, Nikon, Canon, Leica, Fujifilm, and Panasonic have models that support signing photos at the moment they are captured, though Nikon's certificate program was suspended in early 2026 after a signing vulnerability was discovered and has not been restored since.

If an image has Content Credentials attached, you can often view them through the image file itself or a dedicated verification site, which will show you the capture device, any edits made, and whether AI tools touched it along the way.

A practical checking routine

Given the tools that actually exist right now, here is a realistic order of operations when you are unsure about a photo:

The takeaway

The honest state of things in late 2026 is that no single tool can tell you with certainty whether any given photo is AI-generated. SynthID only catches Google's own AI output, Content Credentials only exist where the camera or platform bothered to attach them, and visual tells are fading as models improve. The realistic skill is not spotting a fake on sight, it is knowing which checks exist, running more than one of them, and staying comfortable saying "I cannot verify this" when none of them give you a clear answer.