Skip to Content

How AI could change catfishing

Catfishing has existed in some form or another for centuries. Cyrano de Bergerac anyone?

But, as with everything else, technology has taken catfishing to new depths. We’ve all heard horror stories from dating websites or social media where a person thought they were communicating with was a dashing and successful millionaire instead of a high-school dropout living in his grandmother’s basement.

Now artificial intelligence (AI) is giving catfishing grifters better tools to pull off their cons. Luckily, we have some tips on how to avoid getting catfished.

AI makes catfishing more realistic

Catfishing is an unusual con. Some catfishing can entail cruelly playing with another person’s emotions just because they can.

The aim of other catfishing is to scam a person out of thousands of dollars in the name of romance. In the early digital age, a catfisher’s main tools were emails and photos of other, usually more attractive, people.

Today, as AI improves, catfishers can create fake photos and even videos. By creating a photo using AI, a catfisher doesn’t have to worry about a tech-savvy mark doing a Google Image Search and finding out that the photo is of an actual person and not the lovely “Paola” from Brazil that he’s fallen for.

The website ThisPersonDoesNotExist creates new faces every time you refresh. It relies on a type of machine learning called Generative Adversarial Networks, or GANs.

This program evaluates a huge dataset of images — in this case, human faces — and “learns” how to generate new ones. Called StyleGAN, the GAN code was released by chipmaker Nvidia last year.

Since StyleGAN code is open source, many other sites are starting to generate fake photos as well. One site that doesn’t is WhichFaceIsReal.

The website says its aim is to help the public understand how AI is literally changing the face of society. It also offers advice on how to spot fake photos.

Spotting a fake photo

The creators of WhichFaceIsReal offer a tutorial that helps to spot the tells of a fake photo. They include:


The current StyleGAN algorithm commonly produces shiny blobs that look like water splotches. They can appear anywhere in the image

Background problems:

Force your eyes away from the face in the photo and pay close attention to the background. This is where you can spot blurring and distortions.


Symmetry is still a challenge for facial generation algorithms. Look for asymmetries in facial hair, different earrings in each ear and different forms of collar or fabric on the left and right side.


Sometimes there will be disconnected strands of hair on the face. Or the hair will be too straight and other times the hair will have a halo effect.

With these tips, can you tell if the photo below from ThisPersonDoesNotExist is fake?

Well, what do you think? If you called fake, you’re right.

The major tells are the blurry background and the person next to her. You can see how that person looks like she was cut and pasted into the “original” photo. And, are those fingers or elf ears coming through the fake sidekick’s hair?

If fake photos aren’t enough, after the StyleGan code was developed, it took just three months to create videos of a person moving and talking based on a single sample image.

But there are at least two tried-and-true low-tech tips to protect you from catfishers: They always have excuses to get out of meeting you in real life and they ask for money.

Despite all this, don’t be afraid to wade into the dating pool. With her new e-book, “7 Steps to Online Dating Success,” Kim will walk you through everything you need to know to get the most out of online dating apps and services — and potentially meet that special someone, safely.

Tap or click here for Kim’s 7 Steps to Online Dating Success e-book.

Komando Community background

Join the Komando Community

Get even more know-how in the Komando Community! Here, you can enjoy The Kim Komando Show on your schedule, read Kim's eBooks for free, ask your tech questions in the Forum — and so much more.

Try it for 30 days