“Any sufficiently advanced technology is indistinguishable from magic.” — Arthur C. Clarke, Third Law
I guess I’m a step ahead of HAL 2000, because I can draw. A little. Although based on some new AI technology I have to think he’s lying.
There are, having hit the public consciousness rather suddenly, at least three AI-powered art programs out there. As I understand it, Google’s Imagen is not quite ready for use by the general public. Midjourney has been in the news because a user (Jason Allen, a video-game designer based in Colorado) used it to generate a picture, heavily modified by him using other tools such as Photoshop, to win an award and $300 in an art contest. Despite that the division he entered was clearly labeled Digital Art, other artists raised a fuss because most of the work had been done by the Midjourney software.
I had a look at the picture. And I was blown away. An AI did this? How? (There is no “why”). Yes, Allen modified and played with the original image, but the skill involved came not from a knowledge of painting and drawing but from revising input text over and over and then applying tools from Gimp or Photoshop or something similar. To paint a similar picture with the kind of detail, shading and color selection shown would have taken an artist using traditional methods weeks, if not longer, to complete.
I decided to give Midjourney a try. It’s plainly wonderful, but a bit complicated, and you have to join the Midjourney Discord. As someone who throughout his life is not fond of being compelled to join anything, I dropped it. I then decided to try DALL-E, bearing in mind that it is still in beta and was only open to participants via invitation (a restriction quite likely dropped and open to all by now). Although it lacks basic editing tools, there are other resources available that allow you to work further with the finished generated images.
Enter your text and four images appear within seconds. Four. In seconds. The first try, I must have just stared at the results for an absurd length of time. As I said before: How? After overcoming my initial amazement, I began to work with different images: trying variations (pick one and you get four new images based on it) and more importantly, playing with and modifying the motivating text. At first I just entered random lines that seemed likely to produce interesting outcomes. Think something along the lines of “a stoned Statue of Liberty holding a giant donut instead of a torch, on a cloudy summer day as photographed by Rolling Stone magazine.” Or in the style of Picasso, or whatever tickles your mental fancy.
No, I didn’t do that one. But you are welcome to.
It began to occur to me that I could try to do illustrations to some of my books and stories. This took a lot more work and many more tries to get something specific and, more importantly, accurate. “Human scoutship patrolling above the alien jungle forest of Midworld” took at least 20 tries, variations and subsequent adjustments before I had an image I was satisfied with.
While DALL-E is great with landscapes, buildings, and grand scenes, it’s not there as far as rendering humans is concerned. People tend to be distorted, missing fingers, or bulged in the wrong places. To produce a picture called “Mistress of Lightning” not only took more than 20 attempts, but a fair bit of modifying and cleaning up in Gimp to correct errors of anatomy and remove artifacts that had no place at all in the finished image. The result is — well, if I studied painting and drawing for ten years, I’d never be able to duplicate it. But words I do know how to use, and between those, the AI and the finishing touches I could add, I ended up with a picture I’m rather proud of.
Keep in mind that this kind of software is in its infancy, much as word processors were 30 years or so ago. The algorithms that drive these programs will rapidly improve, to the point where I believe that one day anybody will be able to paint anything. The next step after that will be to visualize an image in your mind and have connected firmware reproduce it, instantly, on a screen, on a canvas or floating in the air in front of you. Very soon, I expect, the builders of such software will be able to animate the images one creates. When that happens, the term ‘home movies’ will acquire an entirely new meaning.
Prescott resident Alan Dean Foster is the author of 130 books. Follow him at AlanDeanFoster. com.