How to Spot AI Fake Images, Deepfake Videos and Cloned Voices
A WhatsApp message arrives with a clip of a famous actor promising guaranteed returns from a new investment platform. Or a midnight call brings your nephew's voice, begging for an urgent transfer. The voice is his and the face is the actor's, yet neither may be real. Generation tools are now available to anyone, and scammers were among the first to put them to work.
In short: Detecting AI-generated fake images comes down to three steps: inspect the visual and audio details critically, trace the image's origin with a reverse search on Google Lens and TinEye, then check provenance data such as Content Credentials and Google's SynthID watermark. Never make a money decision before verifying the person through a separate channel.
Why spotting deepfakes is now an everyday skill
Today a phone app and a few photos from your profile can produce a convincing fake. The US Federal Trade Commission (FTC) has warned that a scammer only needs a short audio clip of a family member, often taken from content posted online, to clone that voice with an available program. We covered this risk earlier in Your voice could be your new digital weak spot.
Your eyes alone are no longer enough: obvious flaws such as extra fingers keep disappearing. You need a method, not a single tell.
Visual clues that reveal AI-generated images
Zoom in and check the edges and background first; models get sloppy at the margins. Look for:
- Hands, ears and teeth: fused fingers, mismatched earrings, too many uniform teeth that look like a single strip.
- Text in the image: signs with warped letters or meaningless words, especially in Arabic script.
- Lighting and shadows: a shadow pointing toward the light source, or a reflection in glasses that doesn't match the scene.
- Overly smooth skin: a plastic finish with no pores or natural blemishes.
- Background: melted faces in the crowd, building lines that bend for no reason, objects blending into one another.
These clues help, but they are not proof. Their absence doesn't make an image genuine.
How to spot a deepfake video
Watch the lips against the audio; even a fraction of a second out of sync is a warning sign. Note odd blinking and blurry face edges when the head turns. On a video call, ask the person to turn their head or pass a hand in front of their face; some real-time face-swap tools stumble at exactly those moments.
Signs of a cloned voice
A fake voice is often emotionally flat, with no natural breaths or with odd cuts. Context matters more: urgent money, secrecy, and refusal to let you call back are enough to doubt the call.
Reverse image search: Google Lens and TinEye
Many "fake" images are just old photos out of context. A reverse search exposes that fast:
- Google Lens: in the Google app on Android and iPhone, or through Circle to Search on supported Android phones, search the image and open About this image. It shows when Google first saw the image or similar ones and how it spread across news sites and social media.
- TinEye: a dedicated engine that finds copies of the same picture, even when cropped or lightly edited. Sort results by oldest to see where it first appeared.
If the photo appeared years earlier from another event, you have your answer.
Checking provenance: Content Credentials and SynthID
Content Credentials (C2PA)
The C2PA standard lets cameras, editing software and AI tools attach a digitally signed record to a file showing how it was made and what was changed. You can upload a file to the verify tool at contentcredentials.org to read that record. Screenshots and files re-sent through messaging apps usually lose this data, so try to get the original file, and avoid uploading sensitive or personal images to outside tools.
Google's SynthID
Google embeds an invisible watermark called SynthID in content created or edited by its own tools. In the Gemini app you can upload an image, video or audio file and ask, "Is this AI-generated?" Google's help page lists file limits, including a 100 MB maximum and videos shorter than 90 seconds.
The key limit: SynthID only detects content made or edited with Google's tools. If Gemini says an image wasn't made with Google AI, that doesn't make it real; another tool may have produced it. We looked at Google's tools for spotting AI content in more detail earlier.
| Tool | What it shows | Limits |
|---|---|---|
| Google Lens (About this image) | First appearance and spread | Doesn't judge if it is generated |
| TinEye | Exact and oldest copies | Smaller index than Google |
| Content Credentials | Signed creation and edit history | Lost in screenshots and re-shares |
| SynthID in Gemini | Content from Google tools | Misses other companies' tools |
Verify the source before the tech
The strongest tool is a simple question: who posted this? Check the account's age and history. An ad featuring a celebrity pushing an investment platform or a miracle remedy should also appear on that celebrity's verified account or a known news site. If it doesn't, assume it's fake.
Common scams that use deepfakes
- Fake celebrity ads: a well-known figure promising fast profits or a "miracle" cure, linking to a site that wants your data or money.
- Cloned calls from relatives: a family member's voice asking for money for an accident or bail, insisting you tell no one.
- Boss impersonation: a voice note or video call from "the manager" demanding an urgent payment to a supplier.
- Fake profiles with generated photos: dating or job profiles with faces of people who don't exist.
All are classic phishing with a more persuasive tool.
What to do if you suspect a fake
- Pause and don't act under pressure. Urgency is the scammer's main weapon.
- Hang up and call back yourself on a number you already know, or reach another family member.
- Agree on a family code word only close relatives know, and ask for it in any odd money request.
- Refuse suspicious payment methods: the FTC flags gift cards, cryptocurrency and wire transfers as warning signs.
- Report the content using the platform's report option, and contact local authorities if you lost money.
- Limit the raw material: keep accounts private and avoid posting long voice clips of your children.
FAQ
How can I tell if a photo was made with AI?
Check hands, text and shadows, run a reverse search, and inspect Content Credentials if present.
Is there a website that detects fake images with 100% accuracy?
No. Detectors give probabilities and can be wrong either way.
Can Gemini detect any AI-generated image?
No. It detects the SynthID watermark from Google's tools only. Images from other generators may pass unnoticed.
What should I do if a relative's voice calls asking for money?
End the call and ring them on their known number, or ask for your family code word. Don't send anything until you're sure.
The takeaway: don't ask "does this look real?" Ask "have I been able to verify it?" The habit of a reverse search and a confirmation call protects you better than any detector.
Sources: US Federal Trade Commission (FTC), Google Gemini Help Center, Google DeepMind, Content Authenticity Initiative, Android Central, TinEye.
Because verifying what we see is now part of protecting ourselves and our money, our interest extends from following Arabic websites and specialized blogs, to what Egypt stores, Kuwait stores, vitamin stores and international sites offer, with Mashhor for social media services as a key to growth.
That's it, friends. We hope you found this useful โ
Best wishes from the #Ezznology team.
Join our community on Telegram and our Facebook group, and follow us on Google News.
You may also like:
Google unveils tools to monitor AI content
Your voice could be your new digital weak spot
