What an AI Face Depixelizer Does to Game Characters and Cartoons

The idea of being able to enhance a pixilated image has long captivated people.

I mean, haven’t you ever seen all those TV shows and movies where they’re looking at surveillance footage or an old photograph on the computer?

images used in face depixelizer
Twitter via boredpanda

Zoom In and Enhance!

The detective always commands a forensics expert to zoom in and enhance the blurry or pixilated image. And lo and behold, the images always crisp up nicely.

Fancy, right?

Well, in real life, it doesn’t really work the way that the silver screen would lead you to believe.

There are, of course, forensic algorithms that work well for image reconstruction. And while it allows forensic experts to reconstruct enough details to make a determination sometimes, it’s not going to give you a crystal clear — or very accurate — picture.

As with any technology, it was soon making its rounds on the internet. And when the internet was presented with a face depixelizer, the testing quickly got hilariously out of control. Because, why not?

The depixelizer takes advantage of top-of-the-line technology to take super-pixilated photos and reconstruct them into high-quality portraits of what the AI thinks is right. It’s based on complex algorithms, programming ingenuity, magical spells… okay, I made that last one up.

Anyway, when used in the way intended, it obviously doesn’t work ideally all the time. It’s still a pretty cool technology, though.

The Internet Responds to the Face Depixelizer

It’s even cooler when various internet users decide to see what bizarre ideas the AI will decide to make a reality with its AI brain.

The meat and bones of the tool were created by Sachit Menon, Alex Damian, and colleagues. In other words, they did some fancy work regarding computer vision and pattern recognition.

It was then developer Denis Malimonov who gave it a simple, easier-to-use interface. That’s what made it more accessible to a wider audience.

He based it on the “PULSE: Self-Supervised Photo Upsampling via Latent Space Exploration of Generative Models” study done by Menon and Damian.

How does it work? The AI gets a pixelated image, which it compares to a bunch of proper quality pictures that it also pixelates down to find one that looks almost identical to the original picture. It then adds several filters and adaptations to make it look as close as possible to the original.

After Malimonov posted a tweet about it, people were quick to test it out — and the results are pretty funny.

“Users immediately began testing the neural network using characters from computer games, and began wondering why the algorithms work so poorly,” explained Malimonov. I mean, what did he expect? It’s the internet!

What Happens When You Load the Face Depixelizer With Video Games and Cartoons?

So, without further ado, here are some of the excellent and hilarious results.

low-res face depixelizer
Twitter via boredpanda
low-res face depixelizer
Twitter via boredpanda
low-res face depixelizer
Twitter via boredpanda
low-res face depixelizer
Twitter via boredpanda
low-res face depixelizer
Twitter via boredpanda
low-res face depixelizer
Twitter via boredpanda
low-res face depixelizer
Twitter via boredpanda
low-res face depixelizer
Twitter via boredpanda
low-res face depixelizer
Twitter via boredpanda
low-res face depixelizer
Twitter via boredpanda
low-res face depixelizer
Twitter via boredpanda