Guided A.I. Creative Image Generation

I have always been interested in art. As a child, when I grew up I wanted to be an artist. Once I was in high school I realized most artists do not make much money and most of the ones that do make money only do so posthumously. This self discovery, while emotionally devastating for 15 year old Mike was a pivotal change in my budding career path. It forced me to look into my second passion in life, computers. I thought I found the perfect career, Graphic Design! After college I spent a couple years pursuing my dreams of becoming the best graphic designer I could be. I made many business cards, notepads and postcards. I was combining my two favorite things, computers and art, but it was not satisfying. I wanted more creativity, more interaction, more motion. Right around this time Adobe Flash was starting to make some waves. It blew my mind the things you could do. I set my mind that Flash was the future and I learned all I could about actionscript and before long I was writing scripts of balls bouncing around, simulating physics (very poorly, but amazing for the time). I was of course wrong about Flash, but it re-adjusted my career path to what it is today. Several hundred websites later I was starting to feel unsatisfied again, I didnt have any creative outlet. I had a third re-awakening a few years ago when I started working with webGL and I changed my career path again to specialize in that.

I have had a third interest, A.I./Machine Learning. The thought of a computer doing something "on its own", or doing something completely novel that it was not directly instructed what/how to do it was magical to me. To play with all these new fancy toys required either a large investment of time in learning and or costly infrastructure out of my reach. With the release of the original DALL-E and recently DALL-E2 I have been feverishly waiting for access. I still have not gotten access to DALL-E2, but I did stumble across an amazing new service that has been consuming all my free time, MidJourney.

At first I was thrown off by the service because it only exists as a chat bot on a discord server. This, at first seemed like such a stupid way to offer such an amazing service but I really wanted to "get my feet wet" with A.I. and art so I started playing around with it. After a week of using it I am convinced that the people at MidJourney are geniuses and discord is the perfect platform for such a service.

A little background, MidJourney is very similar to Open AI's DALL-E. You type a text string describing what you want the image to look like and the system tries to generate an image. Where MidJourney is different is, when you give it a text input it generates four different images. You can then pick any of those four images to be the base image of the next set of four images. Sort of like dog breeding. A litter of puppies is born and as a breeder I pick the top of the litter. Most likely picking pups with desirable traits that are more prevalent. Those picked puppies you use to breed the next generation of puppies and you keep repeating this. This is how MidJourney works, you are constantly picking and iterating until you have an image that you want. Where this gets really interesting is the real-time social interaction. Since this is all happening on a discord server as "chat messages" I see what other people are creating. I see their text inputs and I see their creations evolving over time right along side my creations. When I see someone else's creation that is looking really good I will look at what text inputs they used and might borrow some keywords that they are using, or I could start my own branch of off another persons creation.

Here is an example of an image I was working on the other night...

This is the final image that took about 15 iterations to get to this. It used the text input: "The cutest kitten in the world with huge eyes looking up at the camera. Ultra realistic exaggerated perspective backlighting octane render". The very first 4 images that MidJourney produced looked like this..

From this initial image I decided to try out two "branches" the top right and the bottom right.

This is the second generation of the top right base image.

Third Generation.

and the final generation.. Below is the second branch.

and third generation of this branch.

After a couple itterations the cat started tunring into a hairless cat for some reason, so I decided to pick the best image and create a new image based off the hairless cat and in an effort to add some hair back, I told MidJourney to take the base image and turn it into a "Persian cat". Below is the first four results.

I then took the top right version and started iterating on that.

I have spent more time then I like to admit this last week sitting on discord chatting with a bot. This technology has so much potential and is so exciting. I have TONS more images that I have created, you can check them all out on my Instagram, @the_artificial_artist1. Depending on the popularity of this article I may write up additional ones as I explore the boundaries of A.I. art.

Here are some of the high-res versions of these cat images.