Japanese startup generates photorealistic AI-humans

DataGrid can create whole body model images with high resolution & quality using GAN tech

Japanese startup DataGrid, based on the campus of Kyoto University, has developed an AI algorithm that can spit out an endless parade of realistic-appearing humans that don’t exist. In a video demonstrating the tech, the AI humans continuously morph into new people, showing off different poses and looks.

By using a generative adversarial network (GAN), DataGrid says, “it is possible to generate non-existent, whole body model images with high resolution and high quality,” as stated in its press release.

GAN is a system that uses one AI network to constantly test the product created by a second network. In this case, one network generates images, while the second network tries to spot the ones that are computer generated. This information is then used by the first network to improve the quality of the images it creates. Over time, the generator network is able to create images that cannot be distinguished from real people. 

The company expects the images to find a market in advertising, possibly allowing brands to construct the exact type of person they want to show using their product. The next step, according to DataGrid, is to introduce realistic movement in the fake people.