How to Spot AI-Generated Images and Videos: Practical Checks That Still Work
Visual glitches are disappearing fast, so spotting fakes is now more about checking sources and provenance. Here's a step-by-step routine anyone can follow.
To check whether an image or video is AI-generated, don’t rely on your eyes alone. Check who posted it first, run a reverse image search to find the original, look for Content Credentials or a SynthID watermark, and only then look for visual glitches. If you still can’t confirm where it came from, treat it as unverified and don’t share it.
AI image and video tools have improved so quickly that classic tells like extra fingers are no longer reliable. The good news is that a few habits and free tools catch most fakes that matter.
Step 1: Start with the source, not the pixels
Most misleading AI content spreads because people react before checking. Ask:
- Who posted it? A brand-new account, a page that only posts viral clips, or an account with a name like a news outlet but no history is a warning sign.
- Is anyone credible reporting it? A real dramatic event (a disaster, an arrest, a celebrity incident) will be covered by established news organizations within hours.
- Does it push a strong emotion? Outrage, fear and “too cute to be real” are what fake content is designed to trigger.
- Is there a sales pitch? Fake celebrity endorsements for crypto, supplements or giveaways are a common AI scam. If a “celebrity” is selling something in a video, assume it’s fake until the person’s verified account confirms it.
Step 2: Run a reverse image search
A reverse search shows where else an image has appeared, often revealing the original photo, an earlier date or a fact-check.
- On a phone, take a screenshot (for video, screenshot a clear frame).
- On Android, use Circle to Search; on iPhone, open the Google app and tap the Lens icon, then pick the screenshot. Our Circle to Search and Google Lens guide covers both.
- On a computer, right-click the image in Chrome and choose Search with Google Lens, or upload it to Bing Visual Search or TinEye.
- Look for the earliest copy and any fact-check articles. If the “breaking news” photo first appeared years ago or on an AI art account, you have your answer.
In Google results, the About this image panel (from the three-dot menu next to an image) can show when Google first saw it and whether it carries metadata saying it was created or edited with AI.
Step 3: Look for Content Credentials (C2PA)
Content Credentials are tamper-evident labels based on an open standard called C2PA. A camera, phone or editing app can attach a signed record of how an image was made: which device captured it, and whether AI tools generated or edited it. Some recent phones add them automatically; for example, Google’s Pixel 10 camera signs photos, and Samsung labels images edited with its generative AI tools.
To check:
- Look for a small “CR” icon on the image, which some sites and apps display.
- Or download the file and upload it to the free verify tool at contentcredentials.org/verify.
- Read the history: it may say “Created with an AI tool,” list edits, or show the camera that took it.
The big limitation: missing credentials prove nothing. Most images still don’t carry them, and social networks and screenshots often strip metadata. Credentials tell you where a file came from, not whether what it shows is true.
Step 4: Check for Google’s SynthID watermark
Google adds an invisible watermark called SynthID to images and videos made with its AI tools. You can check a file in the Gemini app:
- Open the Gemini app or gemini.google.com while signed in.
- Upload the image or video.
- Ask “Is this AI-generated?” or type
@synthidto start a check.
Google’s help page currently limits uploads to files of 100 MB or less, and videos to under 90 seconds each. The result only tells you whether Google’s AI made or edited it. A clean result doesn’t rule out other AI tools. Details are on Google’s SynthID verification help page.
Step 5: Look closely, but know the limits
Visual tells are getting rarer, but these still catch lower-effort fakes:
| Check | What to look for |
|---|---|
| Text and signs | Garbled letters, nonsense words, logos that are almost right |
| Hands, teeth, ears, jewelry | Merged fingers, mismatched earrings, odd tooth counts |
| Background | Faces in crowds that melt together, warped railings, impossible architecture |
| Light and shadows | Shadows pointing different ways, reflections that don’t match the scene |
| Skin and texture | Overly smooth, glossy “airbrushed” look across everything |
| Video motion | Objects morphing or disappearing, hair and hands flickering, physics that feels off |
| Video speech | Lip movements slightly out of sync, flat or oddly clean audio, no breathing sounds |
| Clip length | Very short clips stitched together with frequent cuts |
None of these is proof. Real photos can look strange, and the best AI images show none of these flaws.
What about AI detector websites?
Online “AI or not” detectors can be a useful extra signal, but don’t treat their percentages as a verdict. They can flag real photos as AI (especially heavily edited or compressed ones), miss new generators they weren’t trained on, and give different answers for the same file. Never accuse a real person of faking something based on a detector score alone.
Platform labels
Major platforms such as YouTube, TikTok, Instagram and Facebook may show labels like “AI info” or “altered or synthetic content” when a creator discloses AI use or the file carries provenance data. Labels help when present, but plenty of AI content isn’t labeled, so absence of a label doesn’t mean it’s real.
Fake audio and video calls
The same technology is used for cloned voices and live deepfake video calls in scams. If a “relative” or “boss” calls with an urgent request for money, hang up and call them back on a number you know. See our guide to AI voice-cloning scams for a full checklist.
Bottom line
Spotting AI fakes is now mostly about provenance: who posted it, where it first appeared and whether it carries Content Credentials or a SynthID watermark. Use visual clues and detectors as supporting evidence only. When you can’t verify something, the safest move is simple: don’t share it.
FAQ
Can I tell if a photo is AI just by looking?
Sometimes, with low-effort fakes. High-quality AI images often have no visible flaws, so checking the source and provenance is more reliable.
Does a missing watermark mean an image is real?
No. Watermarks and credentials only exist if the tool added them and they weren’t stripped. Their absence tells you very little.
Are AI detectors accurate?
They’re inconsistent. Use them as one signal among several, never as proof.
What should I do if I find a fake?
Don’t share it, even to mock it. Report it to the platform, and if it impersonates you or someone you know, use the platform’s impersonation report form.
