A picture is worth a thousand words—that is what is driving the usage of images in our lives. Now AI tools have emerged that can generate images on every thinkable and even unthinkable topic in an almost unlimited variety of styles. I believe that AI-generated images will completely transform graphic design as desktop publishing did with typesetting and digital imaging with reprographics. In a similar vein as typesetting and reprographics transformed and essentially vanished as a profession AI image creation will have wide-ranging effects on visual design.

The basis of AI image creation technology is vast collections of tagged images—which are widely available on the internet nowadays. A neural network is trained on image-tag combinations by adding an increasing level of diffusion (noise). The AI is used to still recognize a certain object in an ever more diffused image until essentially only noise is visible. Creating an image follows that path backward. A user enters keywords for an image to be created. Starting from a random noise image, the AI tries to recognise portions of the images and attributes it has been instructed to find. It gradually alters the patterns when it recognises specific aspects of the desired output until it is a refined image. The image combines the random seed with a blend of the learned images.

No worries if the process sounds peculiar. How AI networks work is often not even understood by the ones programming it. Ultimately the main point a user cares about is how to get good results. The AI image-generating engines work by text prompts so far. Users enter a description or keywords of a planned image, including styles and attributes. The more precise the description, the better the outcome usually, although fewer words can give surprising results to explore and ignite ideas. Typically within seconds, a selection of four images is presented. Each image can be a starting point for further variations.