A.I. and ArtRage

[alkratzer]

This landscape and the portrait posted earlier were created using a number of steps that start with generating an image using machine learning. Basically, computers are trained on 1,000s of images of faces or landscapes and "learn" the key elements of each to be able to generate new unique "photos" of faces or landscapes that never existed in the real world.

The technical term for this type of Artificial Intelligence is called "Generative Adversarial Networks" (GANs in acronym speak) The "generative" part is obvious, but the "adversarial" part is in reference to part of the training process. Two computers are given the same set of images and then the first creates "new" images and the second one looks at the results and accepts or rejects it based on if it was a successful rendition of a face, landscape or whatever. over many passes, Accepted and rejected images "teach" the first computer to get better at generating believable images. Once you have an initial image, it can be adjusted and tweaked on a number of parameters -- "more trees, less trees" or "smile, no smile" and the like -- until you are satisfied with the result.

The second step to the process involves a different Artificial Intelligence method call "Neural Networks" which are attempting to teach computers to "see" as humans do. It is sometimes called "Texture Mapping" because computers don't see depth, just pixel patterns on a flat surface. If you have ever used one of those apps that make your selfie look like Van Gogh painted it, you are using that technology. The computer has mapped numerous examples of an artist's work and then can "transfer" that style onto any other image. You can do this process using many images or even just one. By taking the generated image from step one you can use step 2 to transform its look. By creating custom collages of example textures you can create personal styles. Some of you might notice that the face posted earlier still looked like some of my earlier work. That is because I choose transformation styles that appeal to my sense of color, brush stroke and the like.

Step three involves creating multiple style versions of the same image using step 2 -- one that looks like oil paint, one that looks like pen and ink, one that introduces a random abstract shapes, etc. and then layering them and blending them in a program like Photoshop. In the face example I also added Art Rage pastel embellishments.

Want to play with this stuff yourself? Create your own images, series of steps, textures and the like?

Step 1 ArtBreeder here:
https://www.artbreeder.com

Step 2 Deep Dream Generator here:
https://deepdreamgenerator.com/

I have a one-man show of this work coming up in September. There is a Step Four that uses A.I. to upscale the images to a high resolution for printing on canvas or metal. AI Gigapixel is my software of choice for that. Pictorem in Montreal is the printing house I use -- they are one of the few that offer a high gloss epoxy glazing on canvases as well as a newer dye sublimation print on metal that is fantastic!