Conrad converted into a Google Map, using the Fast Style Transfer GANConrad Quilty-Harper By Conrad Quilty-HarperArtificial intelligence lets programmers control Barack Obama’s face, replicate the styles of painting greats and even turn a picture of a horse into a zebra. Now you can do the same. Launched this week, a new desktop application called Runway makes it is easy for anyone to run their own AI experiments. The code for many AI projects is already freely available on the internet, but using it normally requires a knowledge of programming and access to powerful and expensive computers. Runway is like an App Store for a type of AI called a generative adversarial network (GAN), which is often used to make and manipulate text, images, or videos. No technical knowledge is needed to use the program. Advertisement Using the Adaptive Style Transfer GAN we created a version of the latest New Scientist cover in the style of Norwegian painter Edvard MunchConrad Quilty-Harper Once you’ve downloaded the app, which is still undergoing testing, it lets you select which tool you’d like to try with only a few mouse clicks. For example, you could choose to play around with StyleGAN, a GAN that is used by the website “This Person Does Not Exist” for generating believable images of people who don’t actually exist. With just a couple of clicks you can create your own artificial face and tweak how similar or different each new face it generates should be. You can also guide the algorithm towards different effects, colours or themes. Running a GAN is power hungry, so Runway does all of its processing on servers hosted in the cloud, which users pay 5 cents per minute for. DenseCAP is a GAN which can describe images. It correctly recognised that Conrad was smiling, and could tell when he was holding a mobile phone, even if it got his shirt colour wrong.Conrad Quilty-Harper Dozens of other different GANs are available on Runway, including facial recognition, image labelling, and OpenAI’s super-trolling text-generating AI. The team say they aim to let people create their own training datasets and even combine the inputs and outputs of different models. “The idea was to create tools that we want to use,” says Cristobal Valenzuela, who co-created the app. To access machine learning before Runway you had to go to a code-hosting website and find a way to access powerful computer chips, he says. Read more: Google has created a maths AI that has already proved 1200 theorems More on these topics: artificial intelligence