AI or Real? How to Tell If a Photo Is AI Generated in 2026

A photo shows up in your feed. A shark swimming down a flooded street. A famous singer at an event they never went to. A storm cloud shaped like a face over a city you know. It looks real, and your thumb is already on the share button. But is it a real moment or an AI generated photo?

Stop for a moment. It might be real. It might also be something a computer made in a few seconds.

Most people are worse at spotting fakes than they think. In a large study by Microsoft’s AI for Good Lab, more than 12,500 people made about 287,000 real or AI judgments and were right only 62% of the time. That is only a little better than a coin toss, and the researchers say people probably do even worse against the newest AI tools.

The good news is that you do not need special skills. Six things to look at and four free checks will help you catch many fakes, even though no single step is proof. This guide shows you how to tell if a photo is AI generated, one simple step at a time.

Quick answer: Look at the photo for at least 10 seconds and check the hands, any text, the light and the background. Then ask Gemini or Google Lens “Is this AI generated?” to scan for the SynthID watermark, test it on Content Credentials Verify, and run a reverse image search with Google Lens or TinEye to find where it first appeared. No single check is proof, so use several together.

Why AI photos are so hard to spot now

A few years ago, AI pictures gave themselves away. People had six fingers, signs were full of nonsense letters, and faces looked like plastic. Those days are mostly gone.

New image models draw hands, readable text and natural skin with ease. Google’s Nano Banana image tool and newer versions such as Nano Banana Pro and Nano Banana 2 can turn a simple prompt into a photo that looks like it came from a phone camera. Other tools can do the same.

Here is the most useful finding for everyday users. Researchers at Northwestern University ran an online test with more than 50,000 people. When they saw an image for one second, they spotted AI images 72% of the time. When they looked for ten seconds, that rose to 80%. The researchers put it simply: take a few more seconds and you get much better at it.

So the first rule is easy. Slow down. Then use the checks below.

6 visual signs of an AI generated photo

Zoom in and go through these one by one. You only need a minute.

1. Hands, teeth, ears and jewellery

In the Northwestern study, body mistakes were the easiest kind for people to spot, so check them first. The newest tools make fewer of these mistakes, so clean hands do not prove a photo is real. Count the fingers. Look at where fingers touch a cup, a phone or another hand. Check if both earrings match, if glasses sit on both ears, and if the teeth look like separate teeth or one white strip.

2. Text, logos and signs

Read every bit of text in the picture, especially small text in the background. Shop signs, number plates, book covers, posters and T shirts often show letters that almost make sense but not quite. A brand logo that looks slightly wrong is another clue.

3. Light, shadows and reflections

In many real photos, light comes from one main source, so shadows usually point the same way. Studio lights and camera flash can break this rule, so treat it as a clue, not proof. Check the direction of every shadow. Look at reflections in mirrors, windows, water and sunglasses. If a person stands in front of a mirror and the reflection shows something different, you have a strong sign.

4. The background

AI tools often give the main subject the most detail, so mistakes tend to show up in the background. Look for straight lines that bend, railings that melt into walls, people in a crowd with blurry or twisted faces, and patterns that repeat in an odd way.

5. Things that could not work in real life

This is the hardest sign to catch, so take your time with it. Ask yourself if every object could actually work. A bag strap that passes through a shoulder. A bicycle with no chain. A chair with legs in the wrong places. A door with no handle on a busy shop. The Northwestern researchers found this kind of mistake, things that could not work in real life, the hardest of all to catch.

6. Skin, hair and the “too perfect” look

Older or weaker tools often make smooth, waxy skin with no pores, hair strands that blend into each other, and lighting that looks like a studio even in a casual scene. The newest tools handle skin well, so a natural look is not proof either. Real life is a bit messy. Real photos usually have some grain, a stray hair, a crooked collar or a random stranger in the corner.

Remember that some real photos also look strange. The Microsoft study found that real photos with unusual light or odd looking details were among the hardest to judge. That is why the next four checks matter.

4 free checks for an AI generated photo

7. Scan for SynthID with Gemini or Google Lens

Google adds an invisible watermark called SynthID to images, videos and audio made with its AI tools. The mark survives common edits like cropping and compression. Google says SynthID has now marked more than 100 billion images and videos. You can check for it in the Gemini app:

1. Open gemini.google.com or the Gemini app on your phone.
2. Tap the plus button or Add files and upload the photo.
3. Ask: Was this created or edited by Google AI?
4. Read the answer. Gemini tells you if it found the SynthID watermark in all or part of the image.

You can do the same check in Google Search. Open the image in Google Lens, AI Mode or Circle to Search on Android and ask Is this AI generated? Google checks it for SynthID and tells you what it finds.

Keep a few limits in mind. OpenAI, Kakao and ElevenLabs have started adding SynthID to their own AI content, but Google’s help page says Gemini can currently only recognize content made with Google AI tools. So a “no watermark” answer does not mean the photo is real. Gemini also has a daily limit of about 10 image checks, and files must be under 100 MB. You can check videos under 90 seconds and audio files the same way.

8. Check Content Credentials

Content Credentials are like a nutrition label for a photo. They travel with the file and can show which camera or app made it and whether AI was used. Some phones now add them on their own. The Pixel 10 camera adds Content Credentials to the JPEG photos it takes, and the Google Photos app shows them in the About panel of an image. Samsung started marking photos edited with Galaxy AI on the Galaxy S25 series. Most photos online still have no credentials at all.

To check any image, go to Content Credentials Verify and drop the file onto the page. If credentials exist, you will see the app or device that created it and any AI edits.

One warning. Many social media apps remove this data when you upload a photo. So if Verify finds nothing, it does not prove the photo is real. It only means the label is missing.

9. Run a reverse image search

A reverse image search shows where else the same picture appears online. Use Google Lens, Circle to Search on Android, or TinEye. Look for the oldest copy you can find.

• If the earliest copy comes from an AI art page or a prompt sharing site, you have your answer.
• If a news agency or a known photographer posted it first, it is more likely real.
• If fact checkers have already written about it, their article will show up here too.

10. Check the source and the story

Ask who posted it first and why. A brand new account with a shocking picture and no other proof is a warning sign. Real events usually have many photos and videos from different people at different angles. If only one perfect picture exists for a huge event, be careful. Search the news for the same story before you believe it.

Should you trust AI image detector websites?

Many websites promise to tell you if a picture is AI with a percentage score. They can help as one signal, but they make mistakes in both directions. They sometimes flag real photos with unusual light, and they can miss images from new AI models that they have never seen.

Treat a detector score like a second opinion, not a final answer. Never share a screenshot that says “99% AI” as proof that someone faked a photo. Put it together with what you found in the checks above.

How to spot AI videos

AI video tools have improved fast too, and the same thinking works. Watch the clip more than once and pay attention to these points:

• Objects that appear, disappear or change shape between frames.
• Text on signs or clothes that changes as the camera moves.
• Hands and fingers that bend in odd ways during movement.
• Lips that do not quite match the words.
• Water, hair, smoke or crowds that move in an unnatural way.
• A corner of the frame that looks cropped or blurred, which can hide a watermark.

You can upload short clips to Gemini as well. It checks videos under 90 seconds for the SynthID watermark.

Quick checklist before you share a photo

New rules on AI labels in 2026

Laws are starting to catch up. In the European Union, the transparency rules in Article 50 of the AI Act apply from 2 August 2026. People who use AI to make deepfakes and publish them must say so, with some lighter rules for art and satire. AI tools must also mark their output in a way computers can read, where this is technically possible. Tools that were already on the market before that date get extra time, until 2 December 2026, to add these marks.

This means more photos will carry watermarks and labels over time. But people who want to fool you will not follow the rules, so your own checks still matter.

What to do if you think a photo is fake

• Do not share it, even as a joke. Shares give it more reach.
• If you already shared it, delete the post and tell your friends it was fake.
• Report it to the platform, especially if it targets a real person.
• If a friend or family member shared it, send them a kind private message with what you found.
• Look for a fact check from a trusted news source before you comment.

Frequently asked questions

Can Gemini tell if ChatGPT or Midjourney made an image?

Not with SynthID, at least for now. OpenAI has started adding SynthID to ChatGPT images, but Google says Gemini can currently only recognize SynthID from Google AI tools. Gemini can also show Content Credentials when a file still has them. For everything else, use a reverse image search and a close look at the details.

If a photo has no watermark, is it real?

Not always. Watermarks and labels can be missing because the tool did not add them, because someone took a screenshot, because an app removed the label, or because heavy edits damaged the watermark. Missing labels mean you need to check other ways.

Are AI image detectors 100% accurate?

No tool is perfect. Detectors can wrongly flag real photos and miss a new AI generated photo. Use them as one clue among many.

Is it safe to upload my photos to these tools?

Think before you upload private or personal photos to any online tool. Read the privacy policy first, and use public images or screenshots when you can.

What is the fastest way to check a viral photo?

Look at it for ten seconds, then run a reverse image search. Those two steps take under three minutes and can catch many fakes, though not all.

Final thoughts

AI images are not going away, and many of them are harmless fun. The problem starts when a fake photo spreads as news. A ten second look, a quick SynthID or Content Credentials check and a reverse image search will help you avoid many fakes. No check is perfect, so keep asking questions before you share.

Want to see how AI tools make these pictures? Read our guide on how to create Gemini AI images, try our Nano Banana prompts, or explore the free AI tools for image generation. Once you know how AI tools make images, you will spot an AI generated photo much faster.