Research Engineer Roland Meertens has taken one of our favorite vintage video game consoles to a whole new level. He’s discovered a new way to capture photorealistic images by using AI technology and a Gameboy.

Using a Gameboy camera that was released in 1998, Roland was able to capture images with only .057 megapixels. From there, the screen would automatically resize the images to 190 x 144 pixels and appear in four various shades of greens and grays—the quality was awful, but the images weren’t completely distorted.

Roland was determined to improve the quality of the photos.

In order to accomplish this he had to create two versions of the same image. One image was the actual picture taken with the Gameboy camera and another was one he created entirely in black (four shades to be exact) and white.

From this point he created a third image by changing the pixel intensity so that the images appear to be lightly glowing.

He discovered that by creating these additional images and combining them with the original, it gives us the illusion that there are more colors and refined details, providing a sharper finished product.




AI technology assists in the experiment by recognizing the faces in the images. After the network has processed multiple images of the same face it begins to recognize the face and process the information faster, allowing Roland to gather more images and refine his experiment more quickly.

He trained his network by running several test sets of images. The technology started to recognize faces and managed to create a photorealistic image after every 100 steps. This may not sound like a lot, but when it comes to enhancing imagery this is a clear step in the right direction.

Roland’s experiment is unique in the sense that it takes technology thought to be obsolete to produce an image acceptable under modern standards. The future of AI is vast and it seems that perhaps anything is possible. Even if that means adults are still playing with their favorite childhood toy.

Check out Roland’s complete in-depth experiment here.

[Featured image via Flickr]