How to Tell if a Picture Is AI: 6 Free Checks (2026)

AI pictures are now so good that a fake product photo, a “breaking news” image or a too-perfect dating profile can fool almost anyone at first glance. The good news is that in 2026 you have real tools to check, not just guesswork.

Google, OpenAI and others now hide watermarks and provenance data in their images, and free tools can read them.

This guide shows you how to tell if a picture is AI generated, step by step: check the hidden SynthID watermark with Gemini, read Content Credentials (C2PA), run a reverse image search, use AI detectors the right way, and spot the visual clues that still give fakes away. It also explains what each check can and cannot prove.

TL;DR

Fastest check: upload the image to the Gemini app and ask “Was this made with AI?”. It reads Google’s invisible SynthID watermark and is starting to read Content Credentials too.
Check Content Credentials (C2PA): drop the file into verify.contentauthenticity.org to see which camera or AI tool made it and how it was edited.
Reverse search it: Google Lens or TinEye shows where the image first appeared, which often settles it.
Look closely: hands, text, jewellery, ears, reflections, shadows and backgrounds still give many fakes away.
Remember: no watermark does not mean real. Screenshots and social media strip metadata, and some generators add no marks at all. Combine several checks.

The 6 Ways to Check, Compared

Method	What it tells you	Reliability	Free?
SynthID in Gemini	Whether Google AI (and tools that adopted SynthID) made or edited it	High when found; silent for other tools	Yes
Content Credentials (C2PA)	Which camera or app made it, and edit history	High when present; often stripped	Yes
Reverse image search	Where and when it first appeared online	Very useful for news and scams	Yes
File metadata (EXIF)	Camera model, software, sometimes the AI tool	Easy to fake or remove	Yes
AI image detectors	A probability score	Medium; false positives happen	Mostly free tiers
Visual inspection	Telltale AI mistakes	Getting harder every month	Yes

Quick answer: run Gemini (SynthID) and a Content Credentials check first, because they give real evidence. Then reverse search. Use detectors and visual clues only as supporting signs.

Step 1: Check the SynthID Watermark in Gemini

SynthID is an invisible watermark from Google DeepMind. It is woven into the pixels, so it survives cropping, resizing and light editing much better than metadata. Google says it has watermarked more than 100 billion images and videos, and OpenAI, Kakao and ElevenLabs announced in May 2026 that they are adopting it.

Open the Gemini app (gemini.google.com or the phone app) and sign in.
Upload the image with the + button.
Ask: “Was this image made or edited with AI?” or “Is this AI-generated?”
Read the answer. If SynthID is found, Gemini says the image was created or edited with Google AI (and may say which parts). It is also rolling out Content Credentials checks in the same place.

Google says this verification has been used more than 50 million times and now covers images, video and audio. It is coming to Google Search (Lens, AI Mode, Circle to Search) and to Gemini in Chrome.

Not sure how Gemini compares with other assistants for everyday tasks? See our ChatGPT vs Claude vs Gemini vs Grok comparison.

Important: SynthID only detects tools that use SynthID. If Gemini finds nothing, the image might still come from Midjourney, an open-source model like Flux or Qwen-Image, or any generator that adds no watermark.

Step 2: Read Content Credentials (C2PA)

Content Credentials (the C2PA standard) are a signed record inside the file that says how it was made: which camera or app, which AI tool, and which edits followed. Adobe Firefly, OpenAI’s ChatGPT images and many cameras add them. Google’s Pixel 10 camera adds them to every photo it takes.

Get the original file, not a screenshot. Download it from the source if you can.
Go to verify.contentauthenticity.org (the Content Authenticity Initiative’s free checker).
Drag the image onto the page.
Read the result. It shows the issuer (for example OpenAI, Adobe or a camera maker), whether AI was used, the edit history and whether the signature is valid.

Other ways to read it: the “Content Credentials” or “cr” icon on sites that support it, the free Verify browser extension, and Google Photos, which shows credentials for Pixel photos. Meta has said Instagram will label camera-captured media with Content Credentials.

No credentials found? That is “inconclusive”, not “fake” or “real”. Screenshots, messaging apps and many social networks strip the data when they re-save the file.

Step 3: Run a Reverse Image Search

Finding where an image came from often answers the question faster than any detector.

Google Lens: on a computer, right-click the image in Chrome and choose “Search with Google Lens”, or upload it at images.google.com. On a phone, use Lens or Circle to Search.
TinEye (tineye.com): sort results by “Oldest” to find the first time the image appeared.
Bing Visual Search as a second opinion.

What to look for:

An older original with different details means the new one was edited or regenerated.
The image on an AI art site (prompt galleries, creator pages) shows it was generated.
No results at all for a “major news” photo is a red flag; real news photos usually come from agencies and appear in many places.
Fact-check articles from outlets like Reuters, AP or AFP often show up for viral fakes.

Step 4: Check the File’s Metadata

Metadata (EXIF and XMP) can name the camera, lens, editing app and sometimes the AI generator.

Windows: right-click the file > Properties > Details.
Mac: open in Preview > Tools > Show Inspector > the (i) tab.
Online: any EXIF viewer will show all fields.

Signs to look for: a “Software” or “Creator tool” field naming an AI app, parameters such as “prompt” or “seed” (some local tools like ComfyUI save the whole workflow in PNG files), or a total lack of camera data on an image claimed to be a phone photo.

Metadata is easy to remove or fake, so treat it as a clue, not proof.

Step 5: Use an AI Image Detector (Carefully)

AI detectors (such as Hive, Sightengine, Illuminarty or “AI or Not”) look at pixel patterns and give a probability that an image is AI. They are useful when there is no watermark, but they are not courts of law.

Upload the highest-quality version you have. Compression and screenshots confuse detectors.
Use two or three detectors and compare. If they disagree, the result is unclear.
Read the score as a hint. 95% “AI” on several tools is a strong sign; 60% on one tool means very little.

Do not accuse someone based on a detector alone. Detectors give false positives on heavily edited real photos, HDR phone shots, upscaled images and art. They also miss new generators they were not trained on.

Upscaled or “enhanced” real photos can look AI-made because the upscaler invents detail. If that is what you are checking, our AI image upscaling guide explains what those tools change.

Step 6: Look for Visual Clues

Modern models fix many old mistakes, but zoom in and you will still often find some of these:

Where to look	Common AI tells
Hands and fingers	Extra or merged fingers, odd nails, rings melting into skin
Text and signs	Garbled letters in the background, fake logos, mismatched fonts
Eyes and teeth	Different reflections in each eye, too-perfect or blurred teeth
Ears, earrings, glasses	Mismatched earrings, glasses that merge into hair or skin
Hair and edges	Hair that blends into the background, halos around the subject
Lighting and shadows	Shadows pointing different ways, light sources that do not match
Backgrounds	Warped buildings, people with smeared faces, repeated patterns
Skin and texture	Plastic, airbrushed skin with no pores; overly smooth fabric
Physics and logic	Objects floating, straps that go nowhere, impossible reflections

Also check the context: is the event real, is the weather right for that date, do the clothes, cars and signs fit the place? Many fakes fail the story test before they fail the pixel test.

Which AI Tools Mark Their Images

Tool	SynthID	Content Credentials (C2PA)
Google Gemini / Imagen / Nano Banana	Yes	Rolling out
OpenAI ChatGPT images, API, Codex	Announced May 2026	Yes
Adobe Firefly	Not announced	Yes
Pixel 10 camera (real photos)	n/a	Yes, marks real camera photos
Open-source models run locally (Flux, Qwen-Image, SDXL)	No	Usually no

Claims for other generators change often, so check each tool’s own help pages. Text has watermarks too; we explain how Claude handles that in Claude text watermark explained.

How to Check on Your Phone

Save the image (long-press > Save) rather than taking a screenshot, to keep as much data as possible.
Open the Gemini app, attach the image and ask whether it was made with AI.
Use Circle to Search or Google Lens on Android, or the Google app on iPhone, to reverse search it.
Check Content Credentials by opening verify.contentauthenticity.org in your phone browser and uploading the saved image.

Real-Life Checks: Scams, News and Profiles

A product photo on a shopping site or marketplace

Reverse search it. Fake shops often use AI images that appear nowhere else, or stolen photos from real brands. Look for garbled text on labels and unrealistic perfect lighting. Ask the seller for a photo with a handwritten note and today’s date.

A shocking news or disaster photo

Check whether major agencies (Reuters, AP, AFP) or local news show the same scene. Look for Content Credentials from a news organisation. Search the claim plus “fact check”. Wait an hour before sharing; fakes are often debunked fast.

A dating or social media profile

Reverse search every profile photo. AI faces often have mismatched earrings, odd backgrounds and identical lighting across “different” photos. Ask for a live video call. Never send money to someone you have not met.

A photo of you or someone you know that you did not take

Save the evidence (link, screenshot, date), report it to the platform, and do not share it further. Many platforms have dedicated reporting for AI-generated images of real people.

What These Checks Cannot Prove

No watermark does not mean real. Many generators add none, and metadata is easily stripped.
A watermark does not show who made it. SynthID tells you a SynthID tool was used, not by whom or when.
Real photos can be edited with AI. A real photo with an AI-removed object may show as “edited with AI”, which is true but not the same as fully fake.
Detectors make mistakes both ways, especially on compressed or upscaled images.

The best answer comes from several checks pointing the same way: a watermark or credential, plus the reverse search history, plus the visual and context clues.

My Take

Start with the evidence tools: Gemini for SynthID and the Content Credentials checker. They take a minute and, when they find something, they are far more reliable than any “AI score”. Then reverse search, because knowing where an image came from solves most real-world cases, from fake shops to viral news.

Use detectors and visual clues to support a decision, never as the only proof. And if an image makes you angry or scared, that is exactly when to slow down and check it.

FAQ

How can I tell if a picture is AI generated for free?

Upload it to the Gemini app and ask if it was made with AI, check it at verify.contentauthenticity.org for Content Credentials, and reverse search it with Google Lens or TinEye. All three are free.

Can Google tell if an image is AI?

Yes, for images made with tools that use SynthID. The Gemini app checks for the watermark, and Google is bringing the check to Search and Chrome.

What is SynthID?

An invisible watermark from Google DeepMind embedded in AI-generated images, video, audio and text. It survives common edits like cropping and resizing.

What are Content Credentials?

A signed record (the C2PA standard) inside a file that shows which camera or app made it and how it was edited. Adobe, OpenAI and Pixel 10 cameras add them.

Are AI image detectors accurate?

They are useful hints, not proof. They can flag real edited photos as AI and miss images from newer generators. Use several and combine them with other checks.

Does a screenshot remove the AI watermark?

A screenshot removes Content Credentials and metadata. SynthID is in the pixels, so it often survives screenshots and resizing, though heavy edits can weaken it.

Can ChatGPT tell if an image is AI?

ChatGPT can describe visual clues, but it cannot reliably detect AI images by looking. Use the Gemini SynthID check and a Content Credentials checker for real evidence.

How do I check if a photo is real?

Look for Content Credentials from a camera (Pixel 10 and some pro cameras add them), find the original source with a reverse search, and check that the details fit the place and date.

How to Tell if a Picture Is AI Generated: SynthID, Content Credentials and 6 Checks (2026)