AI

In late 2022, OPENAI introduced a range of artificial intelligence tools for use by the public.
The company allowed access to DALL·E 2 claiming it to be “a new AI system that can create realistic images and art from a description in natural language.”
Some artists were, and still are, horrified. I sympathise with those who have more than strong reservations about AI. There are many scenarios that fill me with an existential dread.  This essay endeavours to explain how and why I began using it.

For centuries painters have utilized technology for their work. Some have been very secretive about these techniques, Vermeer being a good example. Antonie van Leeuwenhoek, who was regarded as the finest maker of lenses in the world, developed an optical device that enabled Vermeer to project three-dimensional forms onto a flat surface.

In later life, as his eyesight deteriorated, Degas experimented with photography, thrilled by the results, seeing its potential for artistic expression that could be extended to painting. Many frowned on this, claiming it a threat to the sanctity of art, a view still held by several teachers when I studied painting.

Despite these existential fears, 20th-century artists accelerated this trend for using modern technology until the current crop collided with AI.

Being opposed to the Neo-Luddites, I have had no reservations about exploring the use of technology. Witnessing the very start of AI generated images, I was immediately excited by its potential. In particular it offered a solution to a problem.

The arrival of OPENAI coincided with a period of ill health. I suffer from ankylosing spondylitis, and other related issues. These had combined to prevent me from walking. I could no longer venture out and take photographs, which I have always used to develop ideas.

As a student, I learned how Matisse was stricken with arthritis. Instead of painting, he began making his famous paper cut-outs. What I did not learn until decades later was that he employed glamorous assistants to help him in the studio. Unfortunately, I was unable to afford a Matisse-like solution. Instead, I wondered, could AI act as a substitute? The short answer to this question was yes. The longer answer involves some explanation, and in this article I will provide an example of how I used the technology allied to some old-school skills.

First, a little personal background. After leaving school, I took a degree in Aeronautical Engineering.

Like any tool, AI requires an understanding not only of its strengths but also its limitations. Although I am convinced these glitches will disappear in future models, earlier versions made laughable mistakes. If you wanted three-legged men or horror-story hands with fingers like mangled bananas, or an arm with a shoe on the end of it, OpenAI DALL·E-2 was the app for you. I should add that I have tried the latest version, which at the time of writing is DALL·E-3. Unfortunately, I was unimpressed, thinking it inferior to other text-to-image generators – more on that later.
However, I wasn’t looking for accuracy. I simply needed to create compositions, poses, scenes, and lighting effects. I could fix the glitches with old-school skills, namely being able to draw human figures. There was also the non-trivial matter of using one’s imagination. The artist needs an idea of the subject matter before those thoughts are translated into words that can be entered into the AI interface.

In early 2023, I used DALL·E-2 to help construct a digital print. I took inspiration from Don Quixote, which I was reading at the time. I also knew of the ballet. So with that in mind I decided to create a scene from an imaginary ballet, which would be set in a desert at night.

I used DALL·E-2 to produce dozens of images of dancers. Most were unusable. But a few passed muster for further work.

Having selected an image, I used a digital pen linked to Photoshop to redraw the dancer, including changing the costume, adding facial features and hands.

I adopted the same approach to produce the female dancer.

For a large sculpture, which I invented for the staging, I began with a small drawing.

Next, I made a macquette from cardboard before photographing it, and then pixel-painting surfaces on to it, also using Photoshop.

I used DALL·E-2 to produce more than fifty images of illuminated windmills, before settling on one.

Having developed these pieces, I was able to work towards a final image.

There were many iterations, meaning more pixel-painting, before committing the image to print, and several proofs were needed before arriving at a satisfactory result.

OPENAI proved to be too restrictive. At that stage of its development every rendered image had an aspect ratio of 1:1. Instead, I chose to investigate Midjourney, soon discovering results that were a great improvement.

However, the user’s input is crucial to achieving satisfactory results. Is it day or night? What is the setting? What is the aspect ratio? Do you wish to employ specific camera angles and lenses? Which palette would you like to use? What clothes are people wearing? Midjourney can makes guesses, but the results are unlikely to be of use. Moreover, Midjourney cannot cope with too many instructions. One has to build pieces of the jigsaw. This perfectly acceptable for someone who enjoys making collages.

The above image was generated using this prompt:

a full-length body image | GENRE neo-noir, black and white | EMOTION: mysterious | SCENE: A field of white Panama hats. Cate Blanchett in a white dress and straw hat | CAMERA MODEL: Canon EOS R5 | CAMERA LENS: Nikon AF-S NIKKOR 24-120mm f/4G ED VR Lens | SPECIAL EFFECTS: ULTRA-detailed, ultra-photorealistic | TAGS: 8k, award-winning photograph, intense, film still cinematic still, cinematic shot, movie shot, footage from movie, action movie footage, movie scene, cinematic movie scene, –ar 4:3 –v 5.2

Note that in the above example, I instructed Midjourney to use Cate Blanchett as a model. Midjourney is better at producing a type of face, and body, if it has a famous person to work with, rather than:

A willowy blonde woman in her early forties with prominent cheekbones and large eyes

Also note, Midjourney has no restrictions on aspect ratio. In this case it was 4:3. Even so, Midjourney will produce unexpected results. Many will prove to be unsatisfactory. Yet this apparent randomness occasionally inspires new directions, which I find exciting, and it remains a far superior tool to OPENAI.

Finally, do not be surprised if your prompt incurs censorship. Clearly, Midjourney will not produce pornographic images. It is also politically sensitive. I will be writing more about this another time. It will also prevent the user from requesting a tasteful artistic nude.  There are ways around this, and I have been surprised by some of the images generated by an innocent prompt.

I really enjoy using AI and I expect to continue using it. Not only does it save me from walking, and with that a degree of physical pain, it gives me a sense of excitement when seeing an idea quickly unfold as a series of images.

To all the naysayers, I can only claim that it works for me, an artist who chooses to embrace new technology as well as practicing traditional skills, like daubing oil paint with hogs hair brushes.