Underpainting complete for 'Artificial Intelligence Discovering Perfect Pornography'. This painting follows a few weeks of my first experiments with AI to see if it can help with painting creation.
AI, at time of writing, is remarkably rubbish for any creative task, and I suspect it always will be because it aspires to produce an average. This might be good news for competitive and brilliant artists because so much art by rubbish artists (pretty flower painters! quirky landscape painters!) are this, and AI might out-compete them. Still, perhaps this can't be ignored, as the aesthetically ignorant majority like this sort of thing. Art elitism aside I'm concerned with utility and good art, being creative and expressing my unique personality and world, this is the goal of all artists rather than mere craftspeople, us composers rather than mere performers. AI is really powerful and it can have use as an artistic assistant.
I experimented with AI here partly because the painting is about AI, so I thought I'd share a walk though of the process. Here's the idea sketch from around 2014. It was then named 'Alchemist Discovering Perfect Pornography':
I thought the idea was amusing, but, as I've said elsewhere, it seemed to resonate today with the joke of AI 'becoming human'. My paintings always begin with such idea sketches; small, unconsciously sketched ideas which often jump into my head complete with title. These are notes of the image in my head at the time, so often come (in my head) complete with colours and textures. These sketches serve as a mnemonic of that vision. They're more than a seed, but a record of a finished idea which I can revisit when needed.
In the sketch, which I know from that memory but others perhaps can't see, the top 1/3rd of the image is bone, the wiggly blobs on the right are teeth, and the shield-hole above is a skull's nose. Below is a dawn sky over a rocky landscape. The flesh-like rock-like 'monolith' man emerges from it and gazes with fascination at the flower, a human egg, the 'discovery'. The arch in the sky is a reflection of the skull's nose, like a window in the sky, which had (at one point) a sun peeping inside. A falling tooth makes the window something of a mouth of a startled face. At one point (in 2014 when drawing it), I dripped pendulous tears from this window, but erased them.
My goal when realising a painting is to stick to this idea. Sometimes I want to keep the exact shape, and sometimes the shapes are notes for me to find an image (like a rough sketch of a figure that I'd need to locate in an exact pose; I prefer photo-realism, not illustration or cartoon figures). The key goal is to match the feeling. In recent years (well, um ten or so) I've started to blow up the tiny idea sketch (4cm) to full size using tracing paper to closely match the idea's outlines. This is step 1. Then I'll tidy up any rough areas, re-draw and fix up things to look more realistic if needed, like the teeth and skull parts here. I wanted to keep the exact look and shape of the figure here, his expression and pose is vital.
After that, the painting sketch in final size looked like this:
At this point I'd normally locate source materials for the different parts, some skulls, a landscape, a sky. I can invent those things from my mind, like Bob Ross did all of the time, but if I did this my 'happy little' paintings would all look the same and would be based on my same collection of mental images; this would stagnate my art.
One thing I can do is make a lighting model, often from plasticine, and most useful when there are objects which can cast shadows or have complex lighting. I've used computer models before, but they look worse AND take a lot longer than a plasticine model. This, I thought, is an ideal job for AI. It could also help with those textures, saving me trawling free image sites, or the need to photograph real world objects (like myself for the figure, I don't have many skulls hanging around).
I sent the outline drawing to Gemini and described the scene and its textures. To help it, I sent it two images of the texture of bone from my personal image library of texture references. The result wasn't too bad...
...but obviously flawed. The best bit is that bone texture because I gave it some bones, and it placed them well. It didn't get the skull part right, but there's enough there for me to use. Why it added a second head and a second 'flower' I don't know, it absolutely would not stick to my lines. I decided that this didn't matter so much, I'm not using this image for anything except approximate shading.
I wanted things more lit from the 'flower', so I asked it for a sunrise, and to remove the second head and second flower.
Better, more useful (it added a weird hand!). This will do, it's on par with a model lighting guide, with the bonus of more realistic textures and colours. The sky isn't what I had in mind, it's too dark, too cloudy, but I decided to ignore that. It'll never be perfect but I don't want it to be too refined in case it interferes with everything I add as I paint. In this case the actual mistakes and lack of quality help, they leave spaces for me to fill.
The arch window is missing... this was intentional. I decided that, rather than the 'sun of discovery' there, I'd add a couple about to kiss; a robot and a beautiful woman, thus re-inforcing the theme of the painting, so that anyone who had no clue about the title would understand it. I found an appropriate pose from a silent film, and took some images of the robot in Metropolis. These I fed to Gemini and asked it to create the couple.
This worked brilliantly. When supplied with material it's brilliant at extrapolation and transformation, it can change lighting and poses in ways far beyond 2D image editing. Here's the underpainted window:
And the finished underpainting as a whole:
You can see how much of the guide it was or wasn't in the end. It was certainly useful; I've used the bits I wanted, not what I didn't.
I wouldn't consider this an AI artwork at all, but it is AI assisted, particularly the kissing couple. Without the AI I'd use the silent film couple (as I did earlier this year in 'Love And Fragility In The Age Of Perfection') and make the male robotic, but I think this painting is better for that robot. When painting the rest I was constantly aware of not wanting to use the AI imagery, aware that this may dilute my vision, yet some of its suggestions, like the placement of the shadow on the 'wire' were great; this is the sort of thing I'd look to a guide for.
I'd never use AI for writing because no-one could convey what I want to say better than me - that's just obvious. There's no 'better' or 'worse' or 'right' or 'wrong' in that, the best is how we actually write; our personality. This is why I love performing without a script and love improvising on piano; everything we say or do is what our unconscious intended and this is correct, the 'wrong notes' are correct as far as our unconscious creative mind is concerned, correct in an unknown but true way. I've no desire at all to use AI to create music because I love making music with my own tools and I absolutely hate the sound of modern pop (and the idea of pitch correction and the awful sound of auto-tuned vocals!). I've not done any experiments here, maybe it could create a unique sound effect (a Boeing 747 crashing into a junk yard of bedsprings?) or sample that might be useful, or for cleaning up audio, like The Beatles did with 'Now and Then'.
When painting I remained constantly aware of the threat of dilution of the concept by its 'personality'. Perhaps my plasticine lighting guide can also dilute the freedom and impact of that idea sketch. These creative problems are even worse for a film director who has so many competing forces which battle his or her vision. It's a key goal of an artist to weigh up each creative choice and target that intial core vision, that 'orb' of image and emotion which sparked the idea. Perhaps we never succeed, but we can at least be aware of potential pitfalls as we walk that tightrope to our vision, the final artwork.
