Nvidia’s Super Resolution is an AI-powered photo fixing miracle tool that would make CSI jealous

New image enhancement tool finally makes “zoom and enhance” possible.

Even if you’ve never watched any of the CSI TV shows, you’ll probably have a vague recollection of the characters’ ability to spot incredible detail in low-res images. It’s become something of a meme, and you can watch an example below.

What does this have to do with Nvidia? Well, it looks like CSI wasn’t so far off after all (well, not quite as far off).

First, some context. Nvidia is one of the biggest names in AI, with its GPU hardware powering “deep learning” systems from Amazon Alexa to Facebook and pretty much everything in between. And now it’s here to rescue your crappy old low-res photos. Sounds mundane, but it’s actually incredibly clever.

Part of Nvidia’s new and free GameWorks Materials and Tools (GWMT) service, the Super Resolution tool lets you upload any image you want, and within seconds it’ll have produced a better version of that image. This is something of an image editing holy grail.

It’s actually quite amazing, effectively pulling pixels out of thin air. Image enhancement tools of the past have used much more crude methods, simply upping the sharpness or changing contrast and brightness settings in an effort to hide the horrible blurry lines produced when you stretch an image to a size larger than the original.

Related: Best free Photoshop alternatives

Check out the image above and below for examples.

TrustedReviews‘ images sadly can’t quite do justice to the work the tool does, but take it from us that it is impressive. We saw it in action at GDC in San Francisco, and it was able to convert an image in a couple of seconds, making a genuine difference to image quality.

The tool is powered by a deep learning neural network, which is the most common way of training a computer to do a task intelligently without user input. The software behind the service has been fed literally millions of images, a tactic normally used to train self-driving cars and object-recognising robots.

From this, it learns how different shapes and colours interact with each other and can intelligently fill in the missing pixels of an image, producing an outcome for times the size of the original.

TrustedReviews’ home tech editor and TV reviewer Ced Yuen says this technique sounds familiar, but that it’s nonetheless effective:
“Sony has been using this approach in its TVs for years,” he said. “Being able to cross-reference existing images helps to work out how an image should look, and where there is a lack of resolution this is an intelligent way of working out what an image should really look like. It’s not a replacement for a decent source picture, but it’s better than nothing.”

Nvidia’s tool isn’t perfect either, but being able to upscale an image that’s only 256×256 pixels is really quite something. After adding in the missing pixels, you have the basis of an image you can then edit to your own liking, which is much more than you had to start with.

The tools are free, but currently only available to game developers who want to sign up for the beta.