Home / Guides

How to tell if an image is AI-generated

AI images have gotten good enough that "look at the hands" isn't a reliable test anymore. The better approach combines a careful look with provenance checks that tell you where a file came from. Here's how to do both.

Updated October 11, 2026 · By the IsThisAI? team

Quick note: IsThisAI? checks text only. It can't analyze images. This guide points you to the tools and methods that can.

Visual tells

Start by looking closely, at full size if you can. These clues were more reliable a couple of years ago, and newer models make fewer of these mistakes, so treat them as hints:

Text in the image. Signs, labels, book spines, and license plates with garbled or nonsense lettering.
Hands, teeth, ears, and jewelry. Extra or merged fingers, uneven earrings, glasses that melt into skin.
Background logic. Railings that don't connect, windows that don't line up, people in the background with smeared faces.
Lighting and shadows. Shadows pointing in different directions, reflections that don't match the scene.
Too-perfect texture. Skin that looks airbrushed, hair that blends into the background, an overall glossy, illustrated feel.
Context. Does the event make sense? Are there other photos of it from other angles and sources?

A clean-looking image proves nothing, and real photos can look odd because of compression, filters, or phone processing.

Want a quick second opinion on a piece of writing? Paste it into IsThisAI? and see which sentences read as AI-like.

Check your text free →

Metadata and Content Credentials (C2PA)

Content Credentials are a standard way of attaching a verified history to a file. The specification is run by the Coalition for Content Provenance and Authenticity, or C2PA (C2PA). When a file has them, they can show the creator or publisher, when and where it was made, what tools were used, whether generative AI was involved, and edits made along the way. Sites that support them may show a small "CR" icon you can hover over or click (ContentCredentials.org).

To check a file yourself, drag it into the free Content Credentials Verify tool. It accepts common formats including JPEG, PNG, WebP, HEIC, and others. The site itself notes that Content Credentials are still rolling out, so many files won't have any information to show (Content Credentials Verify).

The big catch: metadata can disappear. OpenAI, which attaches C2PA data to images from its tools, says this metadata "can be stripped, lost through uploads and downloads, or broken by transformations like file format changes, resizing, or screenshots" (OpenAI). A screenshot of an AI image usually has no credentials at all. So "no Content Credentials found" means "unknown," not "real."

Invisible watermarks: SynthID and others

Watermarks go a step further by hiding a signal in the pixels themselves, which can survive some edits that strip metadata.

Google SynthID in the Gemini app

Google's SynthID embeds invisible watermarks in content made by Google's AI models. You can upload an image to the Gemini app and ask something like "Is this AI-generated?" or "Was this created or edited by Google AI?" (Gemini Help, Google). How to read the answer, per Google's own help page:

If a SynthID watermark is found, all or part of the image was created or edited by Google AI.
If none is found, it wasn't made or edited by Google AI, but it could have been created by other AI systems.
Some results are unclear, for example very simple images or very small edits.

Google's help page also says Gemini can check Content Credentials, which cover content from many tools (Gemini Help).

OpenAI images

OpenAI says it adds C2PA metadata to images and is also adding Google DeepMind's SynthID watermark to images generated through ChatGPT, Codex, and its API. It's previewing a public tool to check whether an image came from its products, and says that if no signal is found, the tool won't draw a firm conclusion, because signals can be stripped (OpenAI).

The pattern across all of these: a positive result is strong evidence, and a negative result tells you very little.

Reverse image search

Often the fastest way to learn where an image came from is to find where else it appears. Google Lens, Bing Visual Search, and TinEye let you upload an image or paste its link and see matching or similar images online.

Look for the earliest version. If it first appeared on an AI art account or a site that labels generated images, that's your answer.
Check whether news outlets or fact-checkers have covered it.
For a supposed news photo, look for other photos of the same moment from different people. Real events usually produce many angles; a single viral image with no other coverage deserves suspicion.

Tools, and their limits

There are also AI image detectors that try to classify images by their pixels alone. They can be useful as one signal, but they face the same problems as text detectors: they miss things, they flag real photos, and new image models arrive faster than detectors are retrained. We haven't tested image detectors ourselves, so we're not recommending a specific one.

A sensible order of operations:

Look closely for visual tells and check whether the context makes sense.
Run the file through Content Credentials Verify.
Upload it to Gemini to check for SynthID.
Do a reverse image search to find the original source.
If it still matters and you're unsure, don't share it as real.

Checking text instead?

If you're trying to figure out whether a caption, article, or message was written by AI, that's what our detector is for. Start with our guide to the signs of AI-written text, and remember that text detectors have real limits too.

Check any essay, email, or post for AI-sounding writing. Free, no sign-up, and your text stays on your device.

Check your text free →

FAQ

Can IsThisAI? check images?

No. IsThisAI? only checks text. For images, use Content Credentials Verify, the Gemini app's SynthID check, and reverse image search.

How do I check an image for Google's SynthID watermark?

Upload the image to the Gemini app and ask a question like "Is this AI-generated?" or "Was this created or edited by Google AI?" Google says a detected watermark means Google AI created or edited the image; no watermark means it wasn't Google AI, though another AI system could have made it.

What are Content Credentials?

Content Credentials are a provenance standard from the C2PA that can record who made a file, when, with which tools, whether generative AI was used, and what edits were made. You can inspect them with the free Content Credentials Verify tool.

If an image has no Content Credentials, is it real?

Not necessarily. Metadata can be stripped by screenshots, uploads, resizing, or format changes, and many tools don't add it at all. Missing credentials mean the origin is unknown.

More guides

Sources

Coalition for Content Provenance and Authenticity (C2PA). https://c2pa.org/
C2PA / ContentCredentials.org, "Introducing Official Content Credentials Icon". https://contentcredentials.org/introducing-official-content-credentials-icon/
Content Credentials Verify tool. https://verify.contentauthenticity.org/
Gemini Apps Help, "Verify AI-generated images, videos, and audio". https://support.google.com/gemini/answer/16722517?hl=en
Google, "The Gemini app gets new image verification features" (Nov 20, 2025). https://blog.google/innovation-and-ai/products/ai-image-verification-gemini-app/
OpenAI, "Advancing content provenance for a safer, more transparent AI ecosystem". https://openai.com/index/advancing-content-provenance/

This guide is general information, not legal or academic-policy advice. AI detection results, including ours, can be wrong.