I fed excerpts from my novel into an A.I. image generator

The sun is setting on the unintentionally Dadaist / avant-garde phase that has been a hallmark of creative A.I. for several years. We are graduating into an unsettling era of machine learning, where software is eminently capable of generating meaningful art and literature that doesn't trip over it's own digital feet.

Still, at the lower end there are publicly available options such as Craiyon / DALL.E that will draw often surreal images from written prompts. These composites are based on an analysis of millions of images on the Internet along with their associated captions. Evidently the key to defeating Skynet is for everyone to start adding the tag 'Uzi 9mm' to online photographs of kittens.

Recently, I have been using Craiyon to generate artwork for my blog, and I will also be making use of it in my next London, building by building video. I like the imperfect and otherworldly images that it generates. Another point of interest is how easily it is able to render some images while others present a significant challenge. A couple of days ago, it took me half an hour to produce a red macaw wearing a blue neck tie.

In the past, strangers online have occasionally accused me of being an A.I., with one person providing a detailed analysis explaining why this was the case.

Let us assume for a moment that I am not an A.I. but a being of flesh and blood.

I wanted to know how Craiyon would interpret passages from a novel that I published last year, titled 'The Missionary Dune'. How much description could it cope with? What would it focus on and what would it leave out?

The images below are how a non-sentient A.I. perceives a novel written by a sentient human being.

~

image generated by Craiyon
“The reflection of his bright eyes in the counter-top studied the room from underneath the shattered strata of ancient pottery, as if he was a creature at the bottom of a pond, staring upward through the surface film. The ebbing stump of his backcombed quiff registered as a slight, repetitive movement in the glass, keeping perfect time with the small nodding motions of his head, like the minor action of a wristwatch.”

Analysis

Here Craiyon focused very much on the opening lines of the passage. I think the rim at the bottom of the image is an attempt at replicating a pot.

I rather like it. If a stop motion animation studio, in a formerly-Communist Eastern European nation, were to adapt the novel, I think this is what it would probably look like.

~


image generated by Craiyon
“There exists within this collection a peculiar statue, sculpted from white marble that has been contaminated by grey veins of volcanic ash. It depicts a scrum of limbs, belonging to a group of men, who grapple with an enemy in their midst – an emaciated, hollow-eyed figure, with a long, ragged beard. His assailants have succeeded in lifting him off the ground, and it seems as though they are about to bodily cast him out of their presence.”

Analysis

For reasons that remain unclear, this prompt produced pornographic, penis-fixated images of statues, some of which were so obscene that I dared not download them. Of particular note was a visibly aroused statue that appeared to be attempting to force its penis into the mouth of a smaller statue. Another statue had been gifted with a comically gigantic phallus that had blurred in mid-erection. Anyone who is familiar with male arousal will be aware that erections rarely occur fast enough to generate speed lines.

Even this image, which is the best of a bad bunch, has little in common with what is described in the text. Keen observers will note that it seems to depict a man twisting the penis of a reclining figure, while another man looks on with studious interest.

As far as I am concerned, this sexually-sadistic interpretation of a non-sexual excerpt of prose, presents more than adequate grounds for avoiding the current generation of sex bots.

~

image generated by Craiyon
“We first met, by coincidence, on Hampstead Heath in north London, near to the model boating pond. It was a cool Wednesday afternoon in late May. The sun shone weakly overhead. Yusuf was perched on one of the large sections of tree trunk that function as natural benches, having been dragged to their present location from elsewhere in the park. It had rained earlier in the day and the concentric rings of the fallen tree were still damp. A few brown, wind-raked leaves – holdouts from Winter that had refused to decompose and crumble away – clung wetly to the drooping blades of the unmown grass. The gentle downhill slope ended at the fringes of a grey, tarmac footpath that girdled the so-called pond, which would be more accurately described as one of a chain of shallow lakes; if you were to walk all the way around the circumference, I expect that it would take you a good ten minutes. The wind-stippled water retained the shaky, indistinct impressions of the trees that colonised the subdued hills on the far bank, their Spring foliage blotting out any traces of urban London that lay beyond. A sober patch of burnished sun-glare on the mottled surface, gradually shifted position in response to my ever-changing perspective.”

Analysis

Although this is not an accurate depiction of the Hampstead ponds, should one enter the Heath from Merton Lane, it is close enough in spirit. It certainly embodies the curated wilderness of this sizeable green space in Northern London. Also included are dead leaves (rather more than I imagined) and what appears to be “wind-stippled water”. Craiyon did a good job with the tree trunk bench and even included a human figure representing Yusuf al-Makari.

One addition that I strongly approve of, and wish that I had included in the novel, is the living tree emerging from the trunk of a dead tree. Later in the book there are a pair of fir trees that have become conjoined – a mature specimen and a sapling that has taken root in a crook between the massive boughs of its parent. It would have been interesting to exploit these images as a consistent theme throughout the story.

~

image generated by Craiyon
“He was clearly of Arabic descent. It was only later that I learned he was Egyptian by birth. I wrongly judged him to be in his early 30s. In truth, he was much older. He was small and almost birdlike in stature, with the smooth complexion of an adolescent. A crescent-shaped scar, marking his right cheek, anchored its lower tip within the faint semblance of a moustache that negligibly grazed his lip. His short, dark hair was naturally wavy. He had combed it into a severe centre parting, flattening both sides across the scalp. Even with copious gel, he could not prevent it from scrolling at the ends, like an old-fashioned wig.

“He wore a heavy suit jacket, much like my own, tailored from some kind of brown, tweedy material. The rounded tips of a high, open shirt collar were bent backwards by curious design. An olive-green tie, made from rough fabric, was yanked into a tight, scrappy knot, leaving an untidy gap at the neck, where it disappeared between the overlapping halves of a dark, red-trimmed waistcoat.”

Analysis

As you will see below, when it comes to processing character descriptions, Craiyon will deviate between the almost competent and the outright horrifying. It did a good job with Yusuf al-Makari who is the closest thing that the novel has to an antagonist. He is clearly Arabic, with smooth skin and delicate features. I am not sure whether the blemish above the lip is meant to represent his faint moustache or his facial scar, or a combination of both. The hair-style is inaccurate. The clothing is a nod in the right direction. The eyes are unsettling and looked painted-on, but this is par for the course with Craiyon at the moment.

~

image generated by Craiyon
“The room appeared to function both as sleeping quarters, and as a study for Yusuf. The tarnished electric light, given off by the elliptical candle-flame bulbs of a brass chandelier, had bogged down in a gloomy antique clutter of chaotically-arranged furnishings and miscellaneous knick-knacks. The high walls were lined with mismatched, overflowing bookcases that varied tremendously in terms of their age, quality and condition. The books that they held were stacked both vertically and horizontally, and arranged in no particular order. They were mostly poor, battered specimens with bent covers and badly torn and sun-faded dust jackets.

A free-standing, antique basin, with spoked taps, was incongruously plumbed into a shallow alcove in one corner. Thin, vertical strips of the room's former wallpaper protruded from either side of the washstand – a yellow and black floral pattern, the curled-over edges thickly textured with years of accumulated grime. A dulled saucepan skulked in the heavy shadow that pooled where the flared base of the column erupted untidily from the dusty carpet.”

Analysis

The best of many attempts. Overall, I wasn't impressed. This luxurious room, along with its murky furnishings, embodies very little of the seamy, antiquated clutter that Yusuf has built up around himself during his exile in London, though it does incorporate a chandelier. I modelled his living quarters on the front room of a flat in West Hampstead that I used to occasionally visit, and which I recall as being quite large. By accident Craiyon has captured the dimensions of the room and the positioning of the windows rather well, though since none of this features in the description provided, it can hardly be applauded for this achievement.

~

image generated by Craiyon
“...A tattoo of the family coat of arms, on her left shoulder. It consisted of a shield that was divided horizontally across the middle, into green and white sections. The pale top half was occupied by the splayed, symmetrical boughs of a yew tree. The green section below was filled with sinewy silver forms, that I initially mistook for the tree's roots. A closer inspection, undertaken during the drowsy aftermath of our lovemaking, revealed them to be snakes slithering upwards from the pinched tip of the shield and converging around the base of the trunk.”

Analysis

Olinda Munnoch's tattoo of her family coat of arms presented something of a challenge for Craiyon. Most of the time it produced fairly accurate representations of the shield, in terms of it shape and the distribution of the colours, but failed miserably to recreate the heraldic content. On rarer occasions it would take a stab at the design itself while failing to incorporate it into the shield. What Craiyon finally came up with, after many less successful attempts, is better than I expected.

~

image generated by Craiyon
“His mouth had settled into its natural resting expression, stretched slightly, like a rubber band pulled taut between two thumbs. He had very large eyes, like a nocturnal animal, or a bird of prey, set in deep, haggard sockets. The expression that they conveyed fused concern with curiosity, as if he had just caught sight of some potential threat in his peripheral vision. His dark hair was styled like a public schoolboy's, in an unimaginative side parting, that was combed across his scalp, from the left temple, into a gently sloping fringe. He was a tall, wiry man, who appeared to be in the midst of his 30s. There was something almost mantis-like about his poise; the way that he kept his elbows pinned against his sides, with his forearms bent slightly upward.”

Analysis

It pains me greatly to see Edric Munnoch (my favourite character in the novel) depicted as a horrifying anthropomorphic hedgehog mutant with penguin flippers. The alternatives Craiyon produced were no better and many were far worse.

In terms of appearance, the character was modelled on a minor member of the English aristocracy, who thankfully looks nothing like the image on the above.

I think what is demonstrated here is Craiyon's inability to properly understand comparisons. Edric's harried, hawkish appearance has been interpreted literally by the A.I., which has fused aspects of a bird of prey onto the face of a man. The parting of the hair flows in the wrong direction. To Craiyon's credit, it does get the stretched mouth right and at least one of the arms is more or less correctly posed.

~

image generated by Craiyon
“A statuesque young woman had framed herself demurely between the double doors, tilting her head coquettishly to one side, as if she was waiting to be invited in. Her long hair, parted at the centre, cascaded in honey-blonde waves over the shoulder straps of a clingy, sleeveless red dress, that flattered her hourglass figure, the fabric falling into loose, tapering pleats below her knees, and completely obscuring her feet. Her bare arms clasped a small silver purse in front of her slightly-rounded belly. A sequence of silver charms, dangling from the bracelet on her left wrist, reflected the light from the nearest chandelier.”

Analysis

Craiyon had less trouble coming up with images of Edric's sister, Avelina Munnoch, though it seemed reluctant to replicate her dress. At a glance, this picture looks very much like an existing piece of artwork that has been lightly manipulated. However, an image search on Google yielded nothing similar.

As with her brother, I modelled Avelina's appearance on a minor English aristocrat. Unlike Edric, this is a fairly decent interpretation of her appearance, and one that I am very happy with.

~

image generated by Craiyon
“The dune lay in a shallow, circular pit at the foot of the viewing platform. It was smaller than I had imagined – perhaps fifteen-feet in diameter; a contoured expanse of sand the colour of oxidised blood, bearing faint inscriptions in an alphabet that resembled Arabic.”

Analysis

Craiyon failed to create a satisfying depiction of the Missionary Dune. Most of the time it clumsily layered random characters over desert scenes. The only occasion where it attempted to draw the pit, resulted in a descending trio of nesting circles, mottled with what appeared to be purple algal blooms, superimposed over the rippled grey sand of a lunar landscape.

~

image generated by Craiyon
“The widest section of this paved area was occupied by a Victorian pavilion that resembled a miniature one-storey house, with a dirty, red-tiled roof. A towering copse of London plane trees surrounded the building, shielding it from the heavens with their leafy canopies. An illuminated box-sign identified the venue as The Island Place. Directly above the sign, a near-empty, one-litre Coca-Cola bottle was cradled on its side in the arc of the guttering.”

Analysis

'The Island Place' and the plaza that surrounds it, is based on a real location in South London. This isn't a bad representation of the actual structure, however, the canopies of the London plane trees are much higher and there is no fence around it.

~

image generated by Craiyon
“It felt like the thread of my existence was being unravelled,” he said. “It was like an attempt was being made to remove all trace of me from history. I repeated my name, and the names of my wife and my boys, over and over, between my clenched teeth, so I wouldn't forget who I was. I did that for a long time: Five minutes digging, a fifteen minute break, five minutes on the bucket chain, another break, then back to digging.”

Analysis

I was curious as to how Craiyon would interpret an emotional state. In this instance, the A.I. fixated on those parts of the excerpt relating to the character's family, and generated a multitude of images of Indian / Arabic couples, with the occasional cubist portrait thrown in, presumably to represent the narrator's disassembly and removal from history.

It is rather odd that, in this context, the focus was on couples of eastern ethnicities. I wonder whether the mysticism underlying this passage was taken into account. The character who is telling this part of the story is Australian and a bit of a lad, though Craiyon wasn't to know this.

~

image generated by Craiyon
“In the air above them, five cabinet safes, all differing in terms of size and shape, were being carefully lowered by winches from open windows on the third, fourth, and fifth floors. At the time of my arrival, the winches had paused, leaving their payloads dangling pendulously at different altitudes, like the ornaments on a mobile overhanging a child's crib. The smallest safe was no bigger than an old-fashioned analogue television set. It had rounded corners and was painted a glossy British racing green. The largest was a black iron behemoth, with roughly the same dimensions as a wardrobe. It was embossed with the arcing name of its manufacturer – Huckle – the letters painted over in gold.”

Analysis

This is one of those prompts that I was sure would generate an abundance of great images, and didn't, despite numerous attempts. As I was using direct quotes from a novel, I was unable to alter the text in the hope of achieving better results. In this end this was the best I could come up with – a black and white picture of a cabinet safe with what resembles a clockwork door, and an object vaguely reminiscent of a grand piano, dangling from lengths of cable. The other images were either grounded safes, or cables supporting weird abstract objects.

~

image generated by Craiyon
“My dreams were filled with deserts, and with the plaintive roar of a disembodied ocean. I awoke at one point and was momentarily disorientated by the position of the stars in the night sky, as if I had expected them to appear in a different configuration.”

Analysis

If there is one area where Craiyon excels, it is in generating the kind of artificial landscape that one might encounter in poster form, in a new age shop.

This is nothing like I imagined it, but the basic elements are all there, though it did go a bit over the top with the nebula.



image generated by Craiyon
“The shallow pool that filled a marshland depression, inches from the toes of his green wellington boots, held a gaseous reflection of greying, sheep-like clouds, tightly herded together, as they raced past on an airstream that was barely noticeable at ground level. Beneath the flat surface, silty clods of disintegrating soil, strewn across the bottom, were slowly browning the translucent water.

In the middle distance, runic puddles lay scattered like amorphous jigsaw pieces, their faces blank and mercurial among the clumpy, olive-coloured islands that riddled the waterlogged terrain.”

Analysis

When it comes to generating outdoor landscapes, you can be fairly confident that Craiyon will come up with something that is at least plausible. Here the A.I. ignored the Wellington boots entirely, but did good job of rendering the Saltmoor with its misshapen puddles and hazy reflections of the sky.

~

image generated by Craiyon
“On the foreshore, some workers from the estate were dragging sun-blushed sheets of salt from the tidal wash, where they bobbed and jostled against each other like a jagged ice floe.”

Analysis

I have noticed that Craiyon will often ignore colour-cues when they are inconvenient. It is possible here that the A.I. failed to recognise “sun-blushed” as a synonym for rose-coloured.

This prompt produced a host of black and white images, alongside a few that were rather drab and muted in terms of colour. It does, however, convey the physical toil of the estate labourers as they rake the great sheets of sweetheart salt out from the embrace of the tide.

~

image generated by Craiyon
This is how Craiyon imagines Sam Redlark. It falls rather wide of the mark. My forehead horn is on the left side of my face.

The Missionary Dune is available internationally through Amazon as an E-book for the Kindle or as a print on demand paperback.


Comments

Popular posts from this blog

Notes & Queries response: Why do Americans use the term ‘Victorian’?

Notes & Queries response - How different are modern humans from the first Homo sapiens?

Notes & Queries response: How did salt and pepper become the standard table seasonings?