AI & technology Work & ambitions

How to spot a deepfake or AI-generated image

Check where an image, video or voice came from before you believe it.

A course on deepfakes and AI-generated media. It covers how fakes are made, the visual and audio flaws to look for and why they are fading, checking the source and context, finding earlier copies of an image, and verifying a voice or video call through a channel you already trust.

Start with three things you can do today

Type this when Cadence asks what you want to learn

I keep seeing images and videos online that turn out to be AI-generated, and I have heard about cloned voices used in scams. I want to know how deepfakes are made, what gives them away, and how to check before I believe or share something.

Change any word so it fits your situation. The course is built from what you type and a few setup questions.

Start today

Four things you can do before the course exists.

Useful on their own. The course is for when you want the whole skill, not just the first move.

Look for flaws, but do not stop there
The FBI lists signs such as distorted hands or feet, unrealistic teeth or eyes, irregular faces, odd accessories, inaccurate shadows and unrealistic movement. Zoom in on hands, text and jewellery. A clean result does not prove an image is real, because newer tools make fewer of these errors.
Find the first source
Ask who posted it first and whether reliable news outlets report the same event. Run a reverse image search to find earlier copies. Original example: a 'flood photo from today' that also appears in an article from 2019 is old and out of context, even if no AI was involved.
Hang up and call back on a known number
The FTC warns that scammers can clone a loved one's voice from a short clip posted online. If a familiar voice asks urgently for money, hang up and call the person on a number you already have, or check with another family member. Requests for wire transfers, gift cards or cryptocurrency are warning signs.
Agree a family code word
The FBI suggests creating a secret word or phrase with your family to verify identity. Original example: a phrase like 'blue teapot' that never appears online. If someone calling in a crisis cannot give it, verify them another way before sending anything.

Sources: FBI IC3: generative AI fraud warning signs and a family secret word; FTC: scammers using cloned voices in family emergency scams. Practical examples are starting points, not promised results.

Best for: AI & technology
Prerequisites: None. Bring the situation you have.
Course length: 7 to 30 lessons, sized to your goal
Lesson pace: About 10 minutes a day
Format: Read it, or play the audio
Access: Available on the App Store, with early features and Android testing access

Before the course is built

Setup asks about your situation first.

Lesson 1 opens on the moment you describe, and the examples use your role. For this request, expect questions like these.

What worries you most: fake news images, scam calls with cloned voices, or fakes of people you know?
Where do you mostly see media: social feeds, messaging apps, or news sites?
Have you been fooled by, or nearly shared, something fake?

Right for now

If any of this sounds familiar, start here.

You shared something fake

And found out when a friend replied with a fact check.

You worry about a cloned voice

A call that sounds like family, asking urgently for money.

You no longer trust any image

Which is its own problem when the real ones matter.

Course arc

Where the course goes, in order.

Cadence writes the lessons for you, so the exact path follows your answers. This is the ground a course built from this sentence covers.

01 / 06

How deepfakes are made

Face swaps, generated images and voice cloning in plain words.

Visual and audio tells

Hands, text, lighting, lip sync and odd pauses, and why they are fading.

Source and context

Who posted it first, when, and who else reports the same thing.

Finding the original

Reverse image search and tracking down earlier versions.

Voice and video scams

Urgent calls from familiar voices and how to verify them.

Not overcorrecting

Dismissing real evidence as fake is a mistake too.

What changes

What you will be able to do, and what you will stop believing.

You will be able to

Know how fakes are made

Understand in plain terms what AI tools can generate and imitate.

Check before believing

Use source, context and search instead of trusting your eyes alone.

Verify a person

Confirm a voice or video call through a channel you already trust.

Ideas this course corrects

You can always spot a fake by looking closely

Visual flaws help, but generators keep improving. Source, context and independent checks are more reliable than your eyes alone.

Real footage always shows what happened

Genuine video can be cropped, slowed or captioned misleadingly. Check the context as well as whether it is authentic.

Course questions

What to know before you start.

How to spot a deepfake or AI-generated image: what can I do today?

Look for flaws, but do not stop there. The FBI lists signs such as distorted hands or feet, unrealistic teeth or eyes, irregular faces, odd accessories, inaccurate shadows and unrealistic movement. Zoom in on hands, text and jewellery. A clean result does not prove an image is real, because newer tools make fewer of these errors. Find the first source. Ask who posted it first and whether reliable news outlets report the same event. Run a reverse image search to find earlier copies. Original example: a 'flood photo from today' that also appears in an article from 2019 is old and out of context, even if no AI was involved. Hang up and call back on a known number. The FTC warns that scammers can clone a loved one's voice from a short clip posted online. If a familiar voice asks urgently for money, hang up and call the person on a number you already have, or check with another family member. Requests for wire transfers, gift cards or cryptocurrency are warning signs. Agree a family code word. The FBI suggests creating a secret word or phrase with your family to verify identity. Original example: a phrase like 'blue teapot' that never appears online. If someone calling in a crisis cannot give it, verify them another way before sending anything.

Is this a fixed course?

No. Cadence builds the course when you type the sentence and answer a few setup questions, so the lessons follow your situation. This page describes what a course built from this request covers. You see the whole path before Lesson 1.

Who is this course for?

Do I need any prior experience?

No. Setup asks what you already know and where you want to use it, and the course starts from there. Bring a real situation if you have one; the examples are built around it.

How long does it take?

Courses run 7 to 30 lessons of about 10 minutes, sized to what you asked for. One lesson a day is the intended pace, and you can go faster. Lessons can be read or played as audio.

Does this course award a certificate or degree?

No. Cadence is for practical self-learning and does not award accredited degrees or professional certification.