It is 18 months since my last essay exploring AI back in 2023 and there have been many developments since then – certainly more and more people, from poets of my acquaintance to businesses, have explored and made use of AI in one form or another and the industry is full of startups offering AI solutions in all sorts of areas such as creating training videos in which you prime the AI with the content you want delivered and the AI concocts slides with a voiceover or even a fully imagined video trainer to articulate your training needs.
Public awareness has continued to grow, the alarm over AI taking away jobs, or taking over the world and eradicating humans is perhaps less hysterical and the debate more focused. One way in which this is happening is that writers and artists have challenged the AI companies for the currently unregulated and voracious use of their (the creatives) material in training the AI’s LLM’s or Large Language Models. These are the vast bodies of existing work, written and visual, that are fed to AI’s and from which they both learn and plagiarise when prompted to generate an image or a piece of writing “in the style of”. When I first started exploring Generative AI, these ethical battlelines were not so apparent. Still, now we must seriously consider the ethical questions raised by how we choose to use AI – especially when we reference existing artwork or literature. You may feel that the damage is already done, the genie already out of the bottle and that there is no point in bolting the stable door after the horse has bolted, but the laws need changing to protect the copywriter issues advanced by creatives and perhaps there will emerge a statute of limitations so that older work, out of copywrite is usable whilst currently, copyrighted material is either out of bounds or attracts a fee for the use of… Below is an illustration I tried and purely for experiment, I asked for it to be “in the style of” Studio Ghibli – a Japanese animation studio based in Koganei, Tokyo. The exact prompt was “Alice in Wonderland at the Court of the Red Queen in the style of studio ghibli directed by Hayao Miyazaki“
As you can see, Alice is there, in a dining room sumptuous enough to be that of the Red Queen but there are no other characters present and the AI (Midjourney) has become fixated on the Fly Agaric toadstools that Alice found the hookah-smoking Caterpillar sitting on. Does this image owe much to Studi Ghibli and Director Hayao Miyazaki – it certainly could be an animation style – I don’t know his work well enough – I just saw the style in a list of things you could prompt AI with and decided to experiment – but the fact that the AI recognised the name of Studio Ghibli means that it was trained, at some point, by looking at the studio’s work…
So I have decided, for three reasons, to be more circumspect about my use of “in the style of”. Firstly out of fairness to current creatives, secondly because it remains as difficult as ever to get AI to produce the image exactly as you have in your mind’s eye, and lastly I have made less use of AI images to accompany poems – in part because the brilliance of them, not only illustrates the poem but threatens to eclipse or distract from it. However, whilst I am a competent artist in some respects, I am not an illustrator with a wide range of drawing skills and so below, I am going to show you the blend of AI-generated image elements and their combination in PhotoShop to arrive at an image I had in mind for a commissioned illustration. My friend Melissa Lemay, is launching an online journal called Collabotature to showcase collaborative poems and works of literature as well as interviews with authors – she sent me her mission statement and gave me carte blanche to produce an illustration for the launch…
My idea is to have a woman absorbed in reading a book with a “thought bubble” rising up into a night sky showing the moon, and a rocket on it’s way to the moon – all inspired by her reading…
To begin with, I decided on a black-and-white illustration with “drawn” elements combined, which made it slightly easier to achieve consistency. Firstly I wanted a young woman but drawn “in the style of” the E.H.Shepard illustration from “Now We Are Six” by A.A. Milne. – or rather, I wanted her in the pose below, which is not quite the same thing…
The results below, despite what I thought to be a very detailed prompt describing the young woman, her clothes and her pose, was not right…
Cetainly I think the style has little to do with A.H.Shepard and in the lower left picture, the young woman, far from being excited, her imagination fired by reading, has fallen asleep! I tried making variations but nothing worked any better so I then decided to try for a picture of a young girl instead and got the result below.
Once again we have a sleeping beauty, but I decided I could accept the top left image. Next I wanted to have a thought bubble form the girl, featuring a rocket to the moon inspired in the girl’s imagination, by her reading. I was remembering the Moon face in the 1902 film A Trip to the Moon (French: Le voyage dans la lune) by pioneer film director Georges Méliès.
And so I used the prompt “Black and white line drawing of moon against a black background in the style of Georges Melies” to obtain this:-
Nothing like George Melies’ image so no qualms about using the quite straight forward “drawing-style” moon. Next to a rocket, and I have always loved the rocket (was it inspired by the German V2 rockets?) from The Adventures of Tintin, the comics series by Belgian cartoonist Hergé.
And so using this prompt “Black and white rocket against a dark grey background in the style of Tintin Explorers on the Moon” I obtained these images from Midjourney.
Once again, I don’t think the similarity of the rockets warrants the term plagiarism so much as “inspired by…” and I was happy to go with the bottom right image.
Now that I had all the elements, it was time to start assembling them in Adobe PhotoShop – this process is essentially like building a collage except that you can go back and work on each layer ad infinitum, resizing, adjusting the tones, cropping etc. even rearranging the order of the layers – so what hides what…
Having decided on the size and proportion of the Background, I placed the girl before adding a Gradient layer for the night sky. Then, having pasted multiple copies of the Mission Statement text, I placed a text layer in white text so that it fades out at the bottom of the picture but is readable against the black of the sky. I then brought the girl to the front again as the white letters were going over her.
The white letters looked too strong so I increased their transparency to tone them down. Next I added the thought bubbles giving them white edges to stand out and then placed the moon into the large bubble.
I could have added the rocket within the thought bubble, like the moon, but I thought it added to the portrayal of the act of imagination if it came from outside the thought bubble – as if it had come from a different bubble perhaps… Lastly, the rocket looked too static and so I added some “motion streaks” to complete the picture.
So there you have it – each element of the picture was produced by Midjourney generative AI, yet I could never have got an AI to see and conform to the design I had in mind and so I had to assemble them in the way I wanted, after the event and this is one way in which I think AI can be used to aid the graphic designer – after all, collage of existing print material is a very old tradition…