What might the future of automatically generated image captions hold? Image understanding illustrates one of the great gaps between research and commercial practice in today’s deep learning world.

In a new paper, Microsoft researchers propose a system -- ObjGAN -- that can generate images from no more than text captions.

That means anyone can sign up and use it. DALL-E is a deep learning image synthesis model that has been trained on hundreds of millions of images pulled from the Internet.